import re
import numpy as np
import pandas as pd
from tqdm import tqdm
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import f1_score,accuracy_score
from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt
import matplotlib.colors as colors
from matplotlib.colors import ListedColormap
import seaborn as sns
from google.colab import drive
drive.mount('/content/drive')
dir='/content/drive/My Drive/wp1'
Mounted at /content/drive
data = pd.read_csv('/content/drive/My Drive/wp1/vsdatas.csv',na_filter = False)
data = data.loc[~data['comb_no'].isin(['S02R04','S02R05','S02R06'])]
# Voice Sauce full feature set
sessions = set(data['comb_no'].unique().tolist())
scores = []
f1 = []
pos_variables = []
neg_variables = []
Y_tests = []
Y_preds= []
for sp in tqdm(sessions):
# define training and test sets
train_sp = list(sessions.difference([sp]))
train_data = data[data['comb_no'].isin(train_sp)]
test_data = data[data['comb_no'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,3:63]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,3:63]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=5000)
lr.fit(x_train_scaled,y_train)
x_test = scaler.transform(x_test)
scores.append(lr.score(x_test,y_test))
y_pred = lr.predict(x_test)
f1.append(f1_score(y_test, y_pred,average='macro'))
Y_tests.append(y_test)
Y_preds.append(y_pred)
idx = np.argsort(lr.coef_[0])
pos_variables.append((x_train.columns[idx[-10:]].tolist(),np.sort(lr.coef_[0][idx[-10:]])))
neg_variables.append((x_train.columns[idx[0:10]].tolist(),np.sort(lr.coef_[0][idx[0:10]])))
0%| | 0/9 [00:00<?, ?it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 22%|██▏ | 2/9 [00:01<00:06, 1.07it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 33%|███▎ | 3/9 [00:02<00:05, 1.05it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 44%|████▍ | 4/9 [00:03<00:04, 1.03it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 56%|█████▌ | 5/9 [00:04<00:03, 1.03it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 67%|██████▋ | 6/9 [00:05<00:02, 1.03it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 78%|███████▊ | 7/9 [00:06<00:01, 1.02it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 89%|████████▉ | 8/9 [00:07<00:00, 1.02it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 100%|██████████| 9/9 [00:08<00:00, 1.03it/s]
print(np.mean(scores))
print(np.mean(f1))
1.0 1.0
from sklearn.metrics import confusion_matrix
cf_matrix = confusion_matrix(y_test, y_pred)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Test confusion_matrix
Text(33.0, 0.5, 'True')
vs_corr = x_train.corr()
vs_corr.to_csv('/content/drive/My Drive/wp1/vs_corr.csv')
fig = plt.figure(figsize=(35, 30))
sns.heatmap(x_train.corr(), annot=True)
<matplotlib.axes._subplots.AxesSubplot at 0x7f972c555040>
fig.savefig('/content/drive/My Drive/wp1/heatmapvs.png', dpi=300)
lr.coef_
array([[ 0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0.82937119,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
3.22181545, 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0.92077348, 0.13083906, 1.22566263,
0. , 0. , 0. , 0. , 0. ,
0. , -0.24455658, 0. , 0. , 0. ,
-0.59971287, 0.59979662, 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ],
[-2.38299682, 0. , 0. , 0. , 0. ,
0. , 0. , -1.41764988, 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , -0.10350418, 0. , -1.26747798,
-2.04877687, 0. , -0.45949992, 0. , 0. ,
-0.22829303, 0. , 0. , 0. , 0. ,
0. , 0. , -0.17591957, 0. , 0. ,
0. , 0. , 0.12121287, 0.43527955, 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , -1.49845313],
[ 0. , 0. , 0. , 0.04442136, 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0.46829361,
0.35462841, -0.34185744, 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ,
0.08781555, 0. , -0.86126197, 0. , -0.25245789,
0. , 0. , 0. , 0. , 0. ,
0. , 0.60606682, 0. , 0. , -0.54538094,
0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ]])
le_name_mapping = dict(zip(le.classes_, le.transform(le.classes_)))
print(le_name_mapping)
{'BRT': 0, 'CRK': 1, 'MOD': 2}
pos_variables
[(['sF0_mean',
'sF1_mean',
'sB1_sd',
'H2H4c_sd',
'sF2_mean',
'sF1_sd',
'H1A1c_mean',
'H1H2c_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0.03679202, 0.04410526, 0.09492785,
0.10199966, 0.57424031, 0.91126031, 1.18632078, 3.35608436])),
(['sF2_mean',
'sF3_mean',
'sF4_mean',
'Energy_mean',
'soe_sd',
'H1A1c_mean',
'sF1_sd',
'H1H2c_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0. ,
0.77279006, 0.83396865, 1.09056348, 1.3301398 , 2.37248835])),
(['soe_sd',
'HNR25_mean',
'HNR15_mean',
'HNR35_mean',
'H2H4c_sd',
'H1A1c_mean',
'sF1_sd',
'H1H2c_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0.34766437,
0.66512387, 0.71290908, 0.98106699, 1.17390979, 2.66727467])),
(['soe_sd',
'HNR25_mean',
'HNR15_mean',
'HNR35_mean',
'H2H4c_sd',
'sF1_sd',
'H1H2c_sd',
'H1A1c_mean',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0.41962102,
0.43672379, 0.83604564, 1.02716555, 1.26035274, 3.06092528])),
(['soe_sd',
'HNR25_mean',
'HNR15_mean',
'HNR35_mean',
'H2H4c_sd',
'sF1_sd',
'H1A1c_mean',
'H1H2c_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0.19113307,
0.43705577, 0.82207957, 0.96090259, 1.22090655, 3.12423882])),
(['soe_sd',
'HNR25_mean',
'HNR15_mean',
'HNR35_mean',
'sF1_sd',
'H2H4c_sd',
'H1A1c_mean',
'H1H2c_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0.46630669,
0.53890247, 0.66361071, 1.17173498, 1.22540066, 2.97878973])),
(['soe_sd',
'HNR25_mean',
'HNR15_mean',
'HNR35_mean',
'H2H4c_sd',
'sF1_sd',
'H1A1c_mean',
'H1H2c_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0.32853838,
0.45726073, 0.62368254, 0.78649689, 1.22481489, 2.89288767])),
(['sF2_mean',
'sF3_mean',
'sF4_mean',
'Energy_mean',
'soe_sd',
'H1A1c_mean',
'H1H2c_sd',
'sF1_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0. ,
0.29712934, 0.71089774, 1.14527205, 1.55254764, 2.94854225])),
(['soe_sd',
'HNR25_mean',
'HNR15_mean',
'HNR35_mean',
'H2H4c_sd',
'sF1_sd',
'H1A1c_mean',
'H1H2c_sd',
'H1A1c_sd',
'sB1_mean'],
array([0. , 0. , 0. , 0. , 0.13083906,
0.59979662, 0.82937119, 0.92077348, 1.22566263, 3.22181545]))]
neg_variables
[(['sF0_sd',
'HNR05_sd',
'sF2_sd',
'H1c_mean',
'sB4_sd',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd'],
array([-0.61422647, -0.44428738, -0.00335479, 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF2_sd',
'HNR05_sd',
'sF0_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd',
'A3c_sd'],
array([-0.35480105, -0.16872438, -0.10346842, 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF0_sd',
'HNR05_sd',
'sB2_sd',
'sF2_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd'],
array([-0.54428829, -0.40220268, -0.0275248 , -0.01733385, 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF0_sd',
'HNR05_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd',
'A3c_sd',
'H2Kc_sd'],
array([-0.49768497, -0.38123454, 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF0_sd',
'HNR05_sd',
'sF2_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd',
'A3c_sd'],
array([-0.65131439, -0.52905544, -0.06197712, 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF0_sd',
'HNR05_sd',
'sF2_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd',
'A3c_sd'],
array([-0.78087661, -0.35507227, -0.00218767, 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF0_sd',
'HNR05_sd',
'sF2_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd',
'A3c_sd'],
array([-0.70376845, -0.48126558, -0.03706922, 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['HNR05_sd',
'sF0_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd',
'A3c_sd',
'H2Kc_sd'],
array([-0.51794893, -0.00111582, 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF0_sd',
'HNR05_sd',
'H1c_mean',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd',
'A2c_sd',
'A3c_sd',
'H2Kc_sd'],
array([-0.59971287, -0.24455658, 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ]))]
# plots sF0
sns.boxplot(x='phonation', y='sF0_sd', data=data)
<matplotlib.axes._subplots.AxesSubplot at 0x7f97292c21f0>
# plots sB1_mean
sns.boxplot(x='phonation', y='sB1_mean', data=data)
<matplotlib.axes._subplots.AxesSubplot at 0x7f9728f10a90>
# plots H1A1c_mean
sns.boxplot(x='phonation', y='H1A1c_mean', data=data)
<matplotlib.axes._subplots.AxesSubplot at 0x7f97292cb8e0>
# plots H1H2c_sd
sns.boxplot(x='phonation', y='H1H2c_sd', data=data)
<matplotlib.axes._subplots.AxesSubplot at 0x7f97292664f0>
data = pd.read_csv('/content/drive/My Drive/wp1/vsdatasint_nosoe.csv',na_filter = False)
data = data.loc[~data['rs'].isin(['S02R04','S02R05','S02R06'])]
data
| participant | Filename | phonation | interval | H1c_mean | H2c_mean | H4c_mean | A1c_mean | A2c_mean | A3c_mean | ... | sF3_sd | sF4_sd | sB1_sd | sB2_sd | sB3_sd | sB4_sd | session | repetition | comb_no | rs | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | P0001 | AC_P0001_S01_R01_BRT_HDM.mat | BRT | a | 16.126811 | 4.443059 | -0.112041 | -11.467267 | -10.094315 | -1.181237 | ... | 290.526258 | 461.096428 | 169.943927 | 236.018655 | 175.241760 | 214.152636 | S01 | R01 | S01R01a | S01R01 |
| 1 | P0001 | AC_P0001_S01_R01_BRT_HDM.mat | BRT | b | 14.922202 | 3.847778 | -3.113911 | -16.233187 | -15.887453 | -8.517749 | ... | 256.240059 | 417.844357 | 133.029644 | 170.857751 | 193.519834 | 176.821342 | S01 | R01 | S01R01b | S01R01 |
| 2 | P0001 | AC_P0001_S01_R01_BRT_HDM.mat | BRT | c | 16.610802 | 5.388827 | -2.530399 | -17.341007 | -15.475406 | -6.873831 | ... | 194.797725 | 447.314141 | 175.989807 | 237.329558 | 174.363235 | 266.465317 | S01 | R01 | S01R01c | S01R01 |
| 3 | P0001 | AC_P0001_S01_R01_BRT_HDM.mat | BRT | d | 18.044823 | 6.894061 | -1.579303 | -14.783139 | -16.234922 | -4.856212 | ... | 267.265355 | 536.993534 | 130.818743 | 207.481458 | 166.493309 | 230.378719 | S01 | R01 | S01R01d | S01R01 |
| 4 | P0001 | AC_P0001_S01_R01_BRT_HDM.mat | BRT | e | 18.169155 | 6.157108 | -0.264474 | -16.344088 | -16.851381 | -7.165918 | ... | 244.130557 | 441.683714 | 194.517503 | 385.874137 | 174.908489 | 254.829214 | S01 | R01 | S01R01e | S01R01 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 697 | P0004 | AC_P0004_S03_R03_MOD_HDM.mat | MOD | b | 15.699015 | 10.152903 | 1.957719 | -7.538163 | -12.340193 | -8.041816 | ... | 315.327874 | 189.636339 | 87.312050 | 251.221982 | 228.000722 | 207.844723 | S03 | R03 | S03R03b | S03R03 |
| 698 | P0004 | AC_P0004_S03_R03_MOD_HDM.mat | MOD | c | 15.535712 | 10.445065 | 2.631730 | -9.619775 | -14.873524 | -7.127254 | ... | 391.665376 | 288.469289 | 129.305318 | 245.397910 | 212.904208 | 220.231715 | S03 | R03 | S03R03c | S03R03 |
| 699 | P0004 | AC_P0004_S03_R03_MOD_HDM.mat | MOD | d | 17.499325 | 12.125758 | 4.156075 | -5.923121 | -12.405755 | -4.190780 | ... | 446.549735 | 227.426830 | 133.816280 | 306.243223 | 222.570174 | 205.347595 | S03 | R03 | S03R03d | S03R03 |
| 700 | P0004 | AC_P0004_S03_R03_MOD_HDM.mat | MOD | e | 15.003134 | 10.775394 | 2.749792 | -7.902518 | -14.636743 | -5.706246 | ... | 366.066255 | 181.500253 | 78.159472 | 307.840448 | 208.304963 | 206.246489 | S03 | R03 | S03R03e | S03R03 |
| 701 | P0004 | AC_P0004_S03_R03_MOD_HDM.mat | MOD | f | 15.130560 | 8.661292 | 1.109664 | -9.725357 | -13.923312 | -6.778042 | ... | 496.550580 | 255.892037 | 74.339756 | 276.879002 | 178.736760 | 195.832857 | S03 | R03 | S03R03f | S03R03 |
648 rows × 66 columns
# Voice Sauce full feature set
sessions = set(data['comb_no'].unique().tolist())
scores = []
f1 = []
for sp in tqdm(sessions):
# define training and test sets
train_sp = list(sessions.difference([sp]))
train_data = data[data['comb_no'].isin(train_sp)]
test_data = data[data['comb_no'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:62]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:62]
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
x_test = scaler.transform(x_test)
y_pred = np.random.randint(0,2,size=len(y_test))
scores.append(accuracy_score(y_test,y_pred))
f1.append(f1_score(y_test, y_pred,average='macro'))
100%|██████████| 54/54 [00:01<00:00, 35.80it/s]
print(np.mean(scores))
print(np.mean(f1))
0.32098765432098764 0.249152425078351
speakers = set(data['participant'].unique().tolist())
scores = []
f1 = []
for sp in tqdm(speakers):
# define training and test sets
train_sp = list(speakers.difference([sp]))
train_data = data[data['participant'].isin(train_sp)]
test_data = data[data['participant'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:62]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:62]
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
x_test = scaler.transform(x_test)
y_pred = np.random.randint(0,2,size=len(y_test))
scores.append(accuracy_score(y_test,y_pred))
f1.append(f1_score(y_test, y_pred,average='macro'))
100%|██████████| 4/4 [00:00<00:00, 68.92it/s]
print(np.mean(scores))
print(np.mean(f1))
0.3395061728395062 0.2713571705106711
Test data is a set of unseen repetitions.
sessions = set(data['rs'].unique().tolist())
train_scores = []
test_scores=[]
f1 = []
pos_variables = []
neg_variables = []
Y_tests = []
Y_preds= []
for sp in tqdm(sessions):
# define training and test sets
train_sp = list(sessions.difference([sp]))
train_data = data[data['rs'].isin(train_sp)]
test_data = data[data['rs'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:62]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:62]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
# Initialize the logistic regression model
# Train the model
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=4000)
lr.fit(x_train_scaled,y_train)
# Measure accuracy
x_test = scaler.transform(x_test)
test_scores.append(lr.score(x_test, y_test))
train_scores.append(accuracy_score(y_train, lr.predict(x_train_scaled)))
y_pred = lr.predict(x_test)
f1.append(f1_score(y_test, y_pred,average='macro'))
Y_tests.extend(y_test)
Y_preds.extend(y_pred)
idx = np.argsort(lr.coef_[0])
pos_variables.append((x_train.columns[idx[-10:]].tolist(),np.sort(lr.coef_[0][idx[-10:]])))
neg_variables.append((x_train.columns[idx[0:10]].tolist(),np.sort(lr.coef_[0][idx[0:10]])))
100%|██████████| 9/9 [00:33<00:00, 3.68s/it]
# report accuracy
print("Train accuracy:", np.round(np.mean(train_scores), 2))
print("Test accuracy:", np.round(np.mean(test_scores), 2))
print(np.mean(f1))
Train accuracy: 1.0 Test accuracy: 0.98 0.9815207181306757
# Make the confusion matrix (aggregated)
from sklearn.metrics import confusion_matrix
cf_matrix = confusion_matrix(Y_tests, Y_preds)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Test confusion_matrix
Text(33.0, 0.5, 'True')
pos_variables
[(['H2H4c_sd',
'sB3_sd',
'H1A2c_sd',
'H1A1c_mean',
'H1A1c_sd',
'H1c_mean',
'sF2_mean',
'H1H2c_mean',
'sF3_mean',
'sB1_mean'],
array([0.33821481, 0.64734821, 1.26605299, 1.35697026, 1.60104101,
1.64399358, 1.77850003, 2.14235698, 2.29168235, 5.45118081])),
(['H2H4c_sd',
'A3c_mean',
'H1A2c_sd',
'H1A1c_mean',
'H1c_mean',
'sF3_mean',
'H1A1c_sd',
'sF2_mean',
'H1H2c_mean',
'sB1_mean'],
array([0.53557605, 0.57432776, 0.87222402, 1.00251353, 1.05124246,
1.46024644, 1.61268541, 1.8184392 , 2.53893084, 5.18329639])),
(['sB3_sd',
'sB2_mean',
'H1A2c_sd',
'H1A1c_mean',
'H1A1c_sd',
'sF2_mean',
'H1c_mean',
'H1H2c_mean',
'sF3_mean',
'sB1_mean'],
array([0.59786665, 0.60267892, 0.61497295, 1.28027395, 1.3216127 ,
1.52103817, 1.66383963, 2.24485004, 2.3602382 , 5.43454531])),
(['A2c_sd',
'H1c_sd',
'sF1_sd',
'H1c_mean',
'H1A1c_sd',
'sF2_mean',
'H1H2c_mean',
'sF3_mean',
'H1A1c_mean',
'sB1_mean'],
array([0.3889873 , 0.62787789, 0.65322069, 1.39515593, 1.5074849 ,
1.65208761, 1.98819408, 2.03065904, 2.03647584, 5.346468 ])),
(['A2c_mean',
'sB3_sd',
'H1A2c_sd',
'H1A1c_mean',
'H1c_mean',
'H1A1c_sd',
'sF2_mean',
'sF3_mean',
'H1H2c_mean',
'sB1_mean'],
array([0.28087061, 0.72243209, 0.72467066, 1.51010519, 1.56919016,
1.61573958, 1.81090583, 1.93701845, 2.08221548, 5.78908454])),
(['H1A2c_sd',
'sB3_sd',
'H2KH5Kc_sd',
'H1c_mean',
'H1A1c_mean',
'H1A1c_sd',
'sF3_mean',
'sF2_mean',
'H1H2c_mean',
'sB1_mean'],
array([0.40203356, 0.47036445, 0.73503582, 0.9906378 , 1.36507961,
1.60116458, 1.82509543, 2.00530925, 2.22604265, 5.97562516])),
(['H2KH5Kc_sd',
'sB3_sd',
'H1A2c_sd',
'H1A1c_mean',
'H1c_mean',
'sF2_mean',
'H1A1c_sd',
'H1H2c_mean',
'sF3_mean',
'sB1_mean'],
array([0.38824453, 0.6666572 , 1.03808146, 1.30066487, 1.43513651,
1.69758501, 1.88653121, 2.17491933, 3.04183655, 5.23008454])),
(['sF1_sd',
'H1A2c_sd',
'A2c_mean',
'H1c_mean',
'H1A1c_mean',
'H1A1c_sd',
'sF3_mean',
'H1H2c_mean',
'sF2_mean',
'sB1_mean'],
array([0.62591546, 0.6922402 , 0.76338977, 1.10618326, 1.23419316,
1.2651024 , 1.66098209, 1.70133635, 2.18985724, 5.633736 ])),
(['A3c_mean',
'sB3_sd',
'H1A2c_sd',
'H1c_mean',
'H1A1c_mean',
'H1A1c_sd',
'sF2_mean',
'sF3_mean',
'H1H2c_mean',
'sB1_mean'],
array([0.4382078 , 0.554246 , 0.89058112, 1.48356542, 1.60127925,
1.84839066, 1.90352645, 2.10620767, 2.27543419, 5.17747616]))]
Test data is a set of data from an unknowm speaker.
speakers = set(data['participant'].unique().tolist())
train_scores = []
test_scores=[]
f1 = []
pos_variables = []
neg_variables = []
Y_tests = []
Y_preds= []
for sp in tqdm(speakers):
# define training and test sets
train_sp = list(speakers.difference([sp]))
train_data = data[data['participant'].isin(train_sp)]
test_data = data[data['participant'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:62]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:62]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
# Initialize the logistic regression model
# Train the model
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=5000)
lr.fit(x_train_scaled,y_train)
# Measure accuracy
x_test = scaler.transform(x_test)
test_scores.append(lr.score(x_test, y_test))
train_scores.append(accuracy_score(y_train, lr.predict(x_train_scaled)))
y_pred = lr.predict(x_test)
f1.append(f1_score(y_test, y_pred,average='macro'))
Y_tests.extend(y_test)
Y_preds.extend(y_pred)
idx = np.argsort(lr.coef_[0])
pos_variables.append((x_train.columns[idx[-10:]].tolist(),np.sort(lr.coef_[0][idx[-10:]])))
neg_variables.append((x_train.columns[idx[0:10]].tolist(),np.sort(lr.coef_[0][idx[0:10]])))
100%|██████████| 4/4 [00:16<00:00, 4.06s/it]
# report accuracy
print("Train accuracy:", np.round(np.mean(train_scores), 2))
print("Test accuracy:", np.round(np.mean(test_scores), 2))
print(test_scores)
print(np.mean(f1))
# Make the confusion matrix
from sklearn.metrics import confusion_matrix
cf_matrix = confusion_matrix(Y_tests, Y_preds)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Train accuracy: 1.0 Test accuracy: 0.79 [0.691358024691358, 0.7962962962962963, 0.7901234567901234, 0.8703703703703703] 0.7780191027654746 Test confusion_matrix
Text(33.0, 0.5, 'True')
fig = plt.figure(figsize=(35, 30))
sns.heatmap(x_train.corr(), annot=True)
<matplotlib.axes._subplots.AxesSubplot at 0x7f7c86a98f40>
pos_variables
[(['H42Kc_mean',
'H1A1c_sd',
'sB3_mean',
'sB4_mean',
'H1H2c_sd',
'H1c_mean',
'sF3_mean',
'H1A1c_mean',
'sB1_mean',
'H1H2c_mean'],
array([0. , 0.29895416, 0.32427622, 0.3486553 , 0.93669685,
1.15786393, 1.25972968, 2.42587295, 3.18245867, 3.74159054])),
(['sB2_mean',
'A2c_sd',
'H1A1c_mean',
'H1c_mean',
'sF3_mean',
'H1A2c_sd',
'H1A1c_sd',
'H1H2c_mean',
'sF2_mean',
'sB1_mean'],
array([0.2594737 , 0.32892525, 0.5252676 , 0.70832616, 0.79253732,
1.40904406, 1.52778722, 1.58966488, 2.2364696 , 5.56998043])),
(['A2c_sd',
'H1H2c_mean',
'sF1_sd',
'sF3_mean',
'H2KH5Kc_sd',
'H2H4c_sd',
'H1A1c_sd',
'H1A1c_mean',
'sF2_mean',
'sB1_mean'],
array([0.45608928, 0.4879206 , 0.50592975, 0.67248856, 0.76841955,
0.87812098, 1.10281836, 1.55576173, 2.00667015, 5.62369812])),
(['A2c_mean',
'H1A2c_sd',
'sB3_sd',
'H1A1c_sd',
'H1A1c_mean',
'H1c_mean',
'H1H2c_mean',
'sF3_mean',
'sF2_mean',
'sB1_mean'],
array([1.00859104, 1.03829649, 1.22074723, 1.24406107, 1.37156108,
1.84466275, 1.84782741, 1.98092558, 2.20504886, 5.45777367]))]
neg_variables
[(['H2H4c_mean',
'sF0_sd',
'soe_mean',
'sF3_sd',
'sF2_sd',
'sB4_sd',
'H1c_sd',
'H2c_sd',
'H4c_sd',
'A1c_sd'],
array([-0.77639458, -0.01953497, 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. ])),
(['sF0_sd',
'HNR35_sd',
'sF2_sd',
'H42Kc_sd',
'sB4_sd',
'sF3_sd',
'H2c_sd',
'HNR05_sd',
'H1A3c_sd',
'soe_mean'],
array([-1.68465746, -0.53243039, -0.38371883, -0.35549934, -0.34272938,
-0.21079478, -0.19438407, -0.10035395, -0.03377943, 0. ])),
(['sF0_sd',
'sF2_sd',
'CPP_mean',
'sF3_sd',
'H2c_sd',
'sF4_sd',
'sB4_sd',
'A3c_sd',
'H2H4c_mean',
'HNR05_sd'],
array([-2.39669621, -1.36824757, -0.83412098, -0.76033971, -0.57101693,
-0.48278051, -0.45680949, -0.3674082 , -0.26311883, -0.18306736])),
(['sF0_sd',
'sF2_sd',
'sF4_sd',
'HNR15_sd',
'sF3_sd',
'sF1_mean',
'sB4_mean',
'sB4_sd',
'H1c_sd',
'H2c_sd'],
array([-2.55120596, -0.91979736, -0.7989198 , -0.50579007, -0.30625185,
-0.15648423, 0. , 0. , 0. , 0. ]))]
# Selecting the best number of PCs for our data
# PCA of all data
x = data.iloc[:,4:62]
scaler = StandardScaler()
x_scaled = scaler.fit_transform(x)
pca = PCA()
pca.fit(x_scaled)
#Basic info
n_components = len(pca.explained_variance_ratio_)
explained_variance = pca.explained_variance_ratio_
cum_explained_variance = np.cumsum(explained_variance)
idx = np.arange(n_components)+1
df_explained_variance = pd.DataFrame([explained_variance, cum_explained_variance],
index=['explained variance', 'cumulative'],
columns=idx).T
mean_explained_variance = df_explained_variance.iloc[:,0].mean() # calculate mean explained variance
print('PCA Overview')
print('='*40)
print("Total: {} components".format(n_components))
print('-'*40)
print('Mean explained variance:', round(mean_explained_variance,3))
print('-'*40)
print(df_explained_variance.head(20))
print('-'*40)
# Get the eigenvalues
print("Eigenvalues:")
print(pca.explained_variance_)
print()
# Get explained variances
print("Variances (Percentage):")
print(pca.explained_variance_ratio_ * 100)
print()
# Make the scree plot
plt.bar(range(1,len(pca.explained_variance_ )+1),pca.explained_variance_ )
plt.plot(np.cumsum(pca.explained_variance_ratio_ * 100),
label="Cumulative Explained Variance")
plt.xlabel("Number of components (Dimensions)")
plt.ylabel("Explained variance (%)")
plt.legend(loc='upper left')
PCA Overview
========================================
Total: 58 components
----------------------------------------
Mean explained variance: 0.017
----------------------------------------
explained variance cumulative
1 0.193386 0.193386
2 0.151633 0.345019
3 0.121759 0.466778
4 0.088967 0.555745
5 0.060086 0.615831
6 0.045391 0.661222
7 0.039815 0.701038
8 0.032293 0.733331
9 0.026174 0.759505
10 0.025290 0.784795
11 0.022069 0.806864
12 0.020158 0.827022
13 0.015878 0.842900
14 0.014960 0.857861
15 0.013641 0.871501
16 0.012542 0.884043
17 0.010144 0.894187
18 0.009099 0.903286
19 0.008706 0.911992
20 0.008048 0.920040
----------------------------------------
Eigenvalues:
[1.12337207e+01 8.80830058e+00 7.07293560e+00 5.16807221e+00
3.49037113e+00 2.63676500e+00 2.31285610e+00 1.87589966e+00
1.52046531e+00 1.46909768e+00 1.28196501e+00 1.17096340e+00
9.22359882e-01 8.69047700e-01 7.92371863e-01 7.28579554e-01
5.89247017e-01 5.28547324e-01 5.05729507e-01 4.67478445e-01
4.21225826e-01 4.00407489e-01 3.62761156e-01 3.29089467e-01
3.23844356e-01 3.06311594e-01 2.85376907e-01 2.58933163e-01
2.40922343e-01 2.24745360e-01 2.05122698e-01 1.61079984e-01
1.54156640e-01 1.44272220e-01 1.12176444e-01 9.28394431e-02
8.43022509e-02 7.93473226e-02 7.25759947e-02 6.66165465e-02
5.54947391e-02 4.99327417e-02 3.59944164e-02 3.35633309e-02
3.17882722e-02 2.88653155e-02 2.43343768e-02 1.97784095e-02
1.59483226e-02 1.34088366e-02 7.30032814e-03 2.35321693e-03
5.07626882e-07 2.75756459e-07 2.64957555e-07 1.91364801e-07
9.48803447e-08 3.47472201e-08]
Variances (Percentage):
[1.93385943e+01 1.51632888e+01 1.21758975e+01 8.89671861e+00
6.00859441e+00 4.53913089e+00 3.98152910e+00 3.22931853e+00
2.61744641e+00 2.52901820e+00 2.20687357e+00 2.01578682e+00
1.58782153e+00 1.49604582e+00 1.36405011e+00 1.25423311e+00
1.01437532e+00 9.09882180e-01 8.70601827e-01 8.04753496e-01
7.25130666e-01 6.89292372e-01 6.24485068e-01 5.66520023e-01
5.57490683e-01 5.27308432e-01 4.91269846e-01 4.45747543e-01
4.14742326e-01 3.86894019e-01 3.53114053e-01 2.77295524e-01
2.65377144e-01 2.48361341e-01 1.93109193e-01 1.59820987e-01
1.45124405e-01 1.36594609e-01 1.24937922e-01 1.14678868e-01
9.55329294e-02 8.59580775e-02 6.19635680e-02 5.77785097e-02
5.47227866e-02 4.96909832e-02 4.18910755e-02 3.40480816e-02
2.74546741e-02 2.30830069e-02 1.25673486e-02 4.05100935e-03
8.73868115e-07 4.74708463e-07 4.56118397e-07 3.29430147e-07
1.63334353e-07 5.98165480e-08]
<matplotlib.legend.Legend at 0x7f7c86b75f70>
The first 10 components together capture about 78.5% variability in the data. First 15 components: 87.2%. First 18 components: 90%.
Now running PCA again with 15 components.
According to Pipeline I and II, four PCA components can achieve very good prediction. We can run PCA again with only 4 / 5 components.
(no cross validation atm)
# Code source: Gaël Varoquaux, modified by Jaques Grobler
# Modified for documentation by Chenzi Xu
# License: BSD 3 clause
# train and test set
sessions = set(data['rs'].unique().tolist())
le = LabelEncoder()
test = data[data['rs'] == 'S03R03'] # use 'S03R03' as a test set
train = data[data['rs'].isin(list(sessions.difference(['S03R03'])))]
x_train = train.iloc[:,4:62]
y_train = train['phonation'].tolist()
y_train = le.fit_transform(y_train)
x_test = test.iloc[:,4:62]
y_test = test['phonation'].tolist()
y_test = le.transform(y_test)
# Construct pipeline
scaler = StandardScaler()
pca = PCA()
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',max_iter=4000)
pipe = Pipeline([("scaler", scaler), ("pca", pca), ("regressor", lr)])
# Parameters of pipelines can be set using '__' separated parameter names:
param_grid = {
"pca__n_components": [1, 2, 3, 4, 5, 6, 8, 10, 15, 20, 30, 40, 50, 58],
"regressor__C": np.logspace(-1, 2, 8),
}
# We use a GridSearchCV to set the dimensionality of the PCA
search = GridSearchCV(pipe, param_grid, n_jobs=2)
search.fit(x_train, y_train)
print("Best parameter (CV score=%0.3f):" % search.best_score_)
print(search.best_params_)
# Plot the PCA spectrum
x_train_scaled = scaler.fit_transform(x_train)
pca.fit(x_train_scaled)
fig, (ax0, ax1) = plt.subplots(nrows=2, sharex=True, figsize=(6, 6))
ax0.plot(
np.arange(1, pca.n_components_ + 1), pca.explained_variance_ratio_, "+", linewidth=2
)
ax0.set_ylabel("PCA explained variance ratio")
# ax0.axvline(
# search.best_estimator_.named_steps["pca"].n_components,
# linestyle=":",
# label="n_components chosen",
# )
# ax0.legend(prop=dict(size=12))
# For each number of components, find the best classifier results
results = pd.DataFrame(search.cv_results_)
components_col = "param_pca__n_components"
best_clfs = results.groupby(components_col).apply(
lambda g: g.nlargest(1, "mean_test_score")
)
best_clfs.plot(
x=components_col, y="mean_test_score", yerr="std_test_score", legend=False, ax=ax1
)
ax1.set_ylabel("Classification accuracy (val)")
ax1.set_xlabel("n_components")
plt.xlim(-1, 60)
plt.tight_layout()
fig.savefig("/content/drive/My Drive/wp1/pca1.png", dpi=300)
plt.show()
Best parameter (CV score=0.932):
{'pca__n_components': 30, 'regressor__C': 0.7196856730011519}
print(pca.explained_variance_ratio_)
[1.96539328e-01 1.52229921e-01 1.25055762e-01 8.76191661e-02 5.89027439e-02 4.50485598e-02 3.85250041e-02 3.23520070e-02 2.64046028e-02 2.57024504e-02 2.18077669e-02 1.99280038e-02 1.59533196e-02 1.48989616e-02 1.32941615e-02 1.23896257e-02 1.05251401e-02 8.84825698e-03 8.45865172e-03 8.03535345e-03 7.12655782e-03 6.63027064e-03 6.07073586e-03 5.69401282e-03 5.51339969e-03 5.13571125e-03 4.63735317e-03 4.25864308e-03 4.02061150e-03 3.59801186e-03 3.38169668e-03 2.73415833e-03 2.59036851e-03 2.39252098e-03 1.92521458e-03 1.53749227e-03 1.40120101e-03 1.30043310e-03 1.19847210e-03 1.01706938e-03 9.38748680e-04 8.41693385e-04 6.01056878e-04 5.55428575e-04 5.29516216e-04 4.57603964e-04 4.07204588e-04 3.31075013e-04 2.67440626e-04 2.30327052e-04 1.19624822e-04 3.75333494e-05 9.19996780e-09 5.21741764e-09 4.90832661e-09 3.55947589e-09 1.80361667e-09 6.65982320e-10]
# train and test set
speakers = set(data['participant'].unique().tolist())
le = LabelEncoder()
test = data[data['participant'] == 'P0003'] # use 'P0003' as a test set
train = data[data['participant'].isin(list(speakers.difference(['P0003'])))]
x_train = train.iloc[:,4:62]
y_train = train['phonation'].tolist()
y_train = le.fit_transform(y_train)
x_test = test.iloc[:,4:62]
y_test = test['phonation'].tolist()
y_test = le.transform(y_test)
# Construct pipeline
scaler = StandardScaler()
pca = PCA()
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',max_iter=4000)
pipe = Pipeline([("scaler", scaler), ("pca", pca), ("regressor", lr)])
# Parameters of pipelines can be set using '__' separated parameter names:
param_grid = {
"pca__n_components": [1, 2, 3, 4, 5, 6, 8, 10, 15, 20, 30, 40, 50, 58],
"regressor__C": np.logspace(-1, 2, 8),
}
# We use a GridSearchCV to set the dimensionality of the PCA
search = GridSearchCV(pipe, param_grid, n_jobs=2)
search.fit(x_train, y_train)
print("Best parameter (CV score=%0.3f):" % search.best_score_)
print(search.best_params_)
# Plot the PCA spectrum
x_train_scaled = scaler.fit_transform(x_train)
pca.fit(x_train_scaled)
fig, (ax0, ax1) = plt.subplots(nrows=2, sharex=True, figsize=(6, 6))
ax0.plot(
np.arange(1, pca.n_components_ + 1), pca.explained_variance_ratio_, "+", linewidth=2
)
ax0.set_ylabel("PCA explained variance ratio")
ax0.axvline(
search.best_estimator_.named_steps["pca"].n_components,
linestyle=":",
label="n_components chosen",
)
ax0.legend(prop=dict(size=12))
# For each number of components, find the best classifier results
results = pd.DataFrame(search.cv_results_)
components_col = "param_pca__n_components"
best_clfs = results.groupby(components_col).apply(
lambda g: g.nlargest(1, "mean_test_score")
)
best_clfs.plot(
x=components_col, y="mean_test_score", yerr="std_test_score", legend=False, ax=ax1
)
ax1.set_ylabel("Classification accuracy (val)")
ax1.set_xlabel("n_components")
plt.xlim(-1, 60)
plt.tight_layout()
fig.savefig("/content/drive/My Drive/wp1/pca2.png", dpi=300)
plt.show()
Best parameter (CV score=0.965):
{'pca__n_components': 50, 'regressor__C': 13.894954943731374}
print(pca.explained_variance_ratio_)
[2.32445492e-01 1.62898770e-01 1.12319447e-01 6.98924766e-02 5.47571409e-02 4.81151014e-02 3.59350387e-02 3.30721420e-02 2.53200885e-02 2.33440123e-02 2.12082138e-02 1.78811426e-02 1.56217859e-02 1.49506269e-02 1.24902218e-02 1.16941667e-02 1.07430394e-02 9.35159944e-03 7.75708766e-03 7.50814267e-03 6.39642884e-03 6.10130968e-03 5.91809023e-03 5.33885075e-03 4.64410847e-03 4.38246602e-03 4.10699455e-03 3.85489821e-03 3.61825985e-03 3.20444048e-03 3.02386679e-03 2.64519804e-03 2.42755831e-03 2.34588118e-03 1.90799886e-03 1.72761367e-03 1.45435041e-03 1.27217946e-03 1.14066779e-03 9.97739277e-04 9.42067831e-04 8.55344175e-04 7.99649064e-04 6.28712852e-04 5.13372578e-04 4.60495596e-04 4.11875936e-04 3.81167050e-04 3.19269691e-04 3.05086869e-04 2.52843102e-04 1.63683542e-04 1.14618464e-04 3.71476257e-05 1.19451644e-08 6.13171142e-09 4.46361150e-09 3.45789155e-09 1.70733379e-09 4.64210640e-10]
sessions = set(data['rs'].unique().tolist())
train_scores = []
test_scores=[]
f1 = []
Y_tests = []
Y_preds= []
PCA_var = []
for sp in tqdm(sessions):
# define training and test sets
train_sp = list(sessions.difference([sp]))
train_data = data[data['rs'].isin(train_sp)]
test_data = data[data['rs'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:62]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:62]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
# Apply PCA
pca = PCA(n_components=5)
x_train_pca = pca.fit_transform(x_train_scaled)
# Initialize the logistic regression model
# Train the model
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=4000)
lr.fit(x_train_pca,y_train)
# Measure accuracy
x_test = scaler.transform(x_test)
x_test_pca = pca.transform(x_test)
test_scores.append(lr.score(x_test_pca, y_test))
train_scores.append(accuracy_score(y_train, lr.predict(x_train_pca)))
y_pred = lr.predict(x_test_pca)
f1.append(f1_score(y_test, y_pred, average='macro'))
PCA_var.append(pca.explained_variance_ratio_.sum())
Y_tests.extend(y_test)
Y_preds.extend(y_pred)
idx = np.argsort(lr.coef_[0])
100%|██████████| 9/9 [00:03<00:00, 2.98it/s]
# pipe method
# sessions = set(data['rs'].unique().tolist())
# train_scores = []
# test_scores=[]
# f1 = []
# Y_tests = []
# Y_preds= []
# PCA_var = []
# for sp in tqdm(sessions):
# # define training and test sets
# train_sp = list(sessions.difference([sp]))
# train_data = data[data['rs'].isin(train_sp)]
# test_data = data[data['rs'] == sp]
# # Encoding of labels
# y_train = train_data['phonation'].tolist()
# le = LabelEncoder()
# y_train = le.fit_transform(y_train)
# y_test = test_data['phonation'].tolist()
# y_test = le.transform(y_test)
# x_train = train_data.iloc[:,4:64]
# x_test = test_data.iloc[:,4:64]
# # scaler = StandardScaler()
# # x_train = scaler.fit_transform(x_train)
# # x_test = scaler.transform(x_test)
# # Construct pipeline
# scaler = StandardScaler()
# pca = PCA(n_components=5)
# lr = LogisticRegression(penalty='l1',solver='saga',
# multi_class='multinomial',C=10, max_iter=4000)
# pipe = Pipeline([("scaler", scaler), ("pca", pca), ("regressor", lr)])
# pipe.fit(x_train, y_train)
# # Measure accuracy
# test_scores.append(pipe.score(x_test, y_test))
# y_pred = pipe.predict(x_test)
# f1.append(f1_score(y_test, y_pred, average='macro'))
# PCA_var.append(pca.explained_variance_ratio_.sum())
# Y_tests.extend(y_test)
# Y_preds.extend(y_pred)
# idx = np.argsort(lr.coef_[0])
100%|██████████| 9/9 [00:02<00:00, 3.56it/s]
# report accuracy
print("Train accuracy:", np.round(np.mean(train_scores), 2))
print("Test accuracy:", np.round(np.mean(test_scores), 2))
print("f1:", np.round(np.mean(f1),2))
print(PCA_var)
print(np.round(np.mean(PCA_var),2))
# Make the confusion matrix
from sklearn.metrics import confusion_matrix
cf_matrix = confusion_matrix(Y_tests, Y_preds)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Train accuracy: 0.91 Test accuracy: 0.9 f1: 0.9 [0.6153772981796026, 0.6186984195862455, 0.6204263350475345, 0.6105112703063174, 0.6143951856273604, 0.618100652321349, 0.6156646251329854, 0.6142838746365893, 0.6203469213450741] 0.62 Test confusion_matrix
Text(33.0, 0.5, 'True')
pca.explained_variance_ratio_
array([0.19284993, 0.15552851, 0.12329177, 0.08707239, 0.05995583])
test_scores
[0.9444444444444444, 0.9305555555555556, 0.9027777777777778, 0.8055555555555556, 0.9444444444444444, 0.9305555555555556, 0.8194444444444444, 0.9027777777777778, 0.8888888888888888]
PCA loadings are the coefficients of the linear combination of the original variables from which the principal components (PCs) are constructed.
# The coefficients (if 15 PCAs? not updated here)
print('Coefficients: \n', lr.coef_)
fig = plt.figure(figsize=(25, 10))
ax = sns.heatmap(pca.components_,
cmap='PiYG',
yticklabels=[ "PCA"+str(x) for x in range(1,16)],
xticklabels=list(x.columns))
#cbar_kws={"orientation": "horizontal"})
#ax.set_aspect("equal")
#fig.savefig("/content/drive/My Drive/wp1/heatmap_pca1bwr.png")
Coefficients: [[-1.20914646 0. 1.37680422 1.20661823 -0.26520259] [ 1.94062541 -0.08979868 -1.60838848 -1.09261645 1.80625749] [ 0. 0.46662016 0. 0. 0. ]]
# The coefficients
print('Coefficients: \n', lr.coef_)
fig = plt.figure(figsize=(25, 5.6))
ax = sns.heatmap(pca.components_,
cmap='PiYG',
norm=colors.Normalize(vmin=-0.32,vmax=0.32),
yticklabels=[ "PCA"+str(x) for x in range(1,6)],
xticklabels=list(x_train.columns))
#cbar_kws={"orientation": "horizontal"})
ax.set_aspect("2.5")
#fig.savefig("/content/drive/My Drive/wp1/heatmap_pca1short.png", dpi=300)
Coefficients: [[-1.20914646 0. 1.37680422 1.20661823 -0.26520259] [ 1.94062541 -0.08979868 -1.60838848 -1.09261645 1.80625749] [ 0. 0.46662016 0. 0. 0. ]]
print(pca.components_)
[[ 2.40285081e-02 1.77082415e-01 1.47603118e-01 1.40939874e-01 5.83809584e-02 -4.01813761e-03 2.50528252e-02 -1.61495091e-01 8.08417565e-02 -1.03261331e-01 -2.75513151e-02 2.39675198e-02 1.23630853e-01 3.41563170e-02 1.55790231e-01 5.58767042e-02 -2.28391266e-01 -1.83665945e-01 -1.68187678e-01 -1.36605898e-01 -1.97831550e-01 -9.50100142e-03 -6.54047644e-02 -1.66316477e-01 -9.65600354e-02 -2.11899490e-01 -6.99901218e-02 -1.20679801e-01 -4.88085620e-02 -2.33471585e-01 -1.79333089e-01 -1.99979040e-01 -2.18093517e-01 -1.88600152e-01 -2.13300442e-01 -2.16183499e-01 -5.26054160e-02 -1.96273804e-01 -1.72762424e-01 -1.44536377e-01 -1.62458738e-01 -1.05475003e-01 -1.43362876e-01 6.95573854e-03 5.81363152e-02 -1.06615168e-01 -4.51750375e-02 -6.13646106e-02 -8.84948647e-02 -3.09451976e-02 -1.94022907e-01 -1.16215477e-02 1.19750961e-02 1.43136729e-02 -1.83600178e-01 -8.34309229e-03 -9.48236798e-03 1.76186243e-02] [ 1.17600357e-01 1.50654728e-01 2.13850966e-01 2.40813247e-01 2.61185783e-01 2.50017525e-01 2.66142304e-01 -1.50292559e-02 -1.20992392e-01 -1.00748531e-01 -1.12801433e-01 -8.10556871e-02 -6.49040732e-02 1.51299706e-01 1.72781975e-01 2.56102601e-01 -6.47618312e-02 -1.33717082e-01 -1.29580395e-01 -1.23400909e-01 8.18165293e-02 -7.50421624e-02 -8.46504809e-02 -4.48581334e-02 9.92970296e-03 -2.19345513e-02 -5.66765472e-02 -6.01421444e-02 -8.53608300e-02 1.22455386e-01 1.55823798e-01 1.78312272e-01 1.43861990e-01 1.71856229e-01 1.34707975e-01 1.24923031e-01 -1.09900196e-02 3.48897704e-02 1.95035362e-02 -1.23997446e-02 -3.25174901e-02 3.22278673e-03 1.92576527e-02 1.71362102e-01 2.52723302e-01 1.62603378e-01 1.75606608e-01 1.87015900e-01 1.74397213e-01 1.15545935e-01 -1.71297832e-02 4.68846597e-03 5.92658860e-03 -1.10645455e-01 -1.48875806e-02 1.84807162e-03 -8.20826861e-03 -2.32665421e-02] [ 3.20804840e-01 1.83245924e-01 1.39665827e-01 2.10649346e-02 3.29970529e-02 -6.06460645e-02 -1.30610355e-02 2.05556485e-01 1.11796020e-01 2.98134874e-01 3.01810423e-01 3.25212176e-01 1.55981860e-01 -1.77455463e-01 -1.73956934e-02 1.53490221e-01 -1.78126241e-01 -1.48215005e-01 -1.58757732e-01 -1.13038643e-01 -1.55566231e-01 6.50044332e-02 6.79619844e-02 -1.03797503e-01 6.75542334e-02 1.64813877e-01 1.54074240e-01 5.00672489e-02 -1.03753545e-02 8.29642118e-02 -9.12161210e-03 -1.98162369e-02 3.47195202e-02 3.32347416e-02 -1.66048573e-02 -2.36233322e-02 8.78009357e-02 -2.91291340e-03 1.82068754e-01 1.85960198e-01 8.14821141e-02 3.84086154e-02 -3.58116091e-02 -6.91253751e-02 1.15036953e-01 -2.12519744e-02 -6.54024429e-03 -1.33841111e-02 -3.51092868e-02 -1.99728188e-01 1.01184779e-01 -9.60186724e-02 1.02672208e-01 1.30861456e-04 1.69623278e-01 6.84935678e-02 7.88416905e-02 -5.68907239e-03] [ 4.11653382e-02 -1.30540166e-01 -6.05575124e-02 -3.62823498e-02 1.82853181e-01 2.12362581e-01 1.89169300e-01 1.93190662e-01 -1.63369926e-01 7.34040154e-02 -1.21495827e-01 -1.20203431e-01 -2.62431555e-01 1.12220734e-01 -2.11670342e-01 -6.36701698e-02 -4.42349779e-02 -1.34185715e-01 -1.93271142e-01 -2.54652238e-01 6.61828812e-02 -1.19398059e-01 -1.49837350e-02 -6.82775605e-02 -2.15179989e-01 1.74840973e-01 -6.45338709e-02 -1.37724171e-01 1.18832115e-01 1.97761624e-02 -1.61850305e-01 -8.83036172e-02 -9.13398677e-02 -3.71965645e-02 -1.68242115e-02 -7.38221617e-02 1.63224320e-01 7.14792930e-02 3.28197615e-02 9.73383413e-02 2.87356436e-02 -1.46204007e-01 -4.36222447e-02 -2.34099250e-01 -6.05910077e-02 -1.60424789e-01 -1.34104350e-01 -9.19729703e-02 -4.37506239e-02 8.01331473e-02 1.40704563e-01 -6.11645595e-02 -1.45418986e-01 2.02629526e-01 1.55482300e-01 -1.22315103e-02 -1.58686526e-01 5.22392591e-02] [-1.14046842e-01 -5.35467977e-02 -1.10615992e-01 -2.13195890e-01 -1.56812031e-01 -1.52151816e-01 -1.22251630e-01 -8.56450678e-02 1.17421560e-01 7.93925628e-02 2.28547619e-02 1.21737953e-02 1.71856393e-02 -4.21999210e-02 2.40600557e-02 -6.28995602e-02 -1.49081640e-02 -1.47832800e-01 -1.26007226e-01 -1.14074338e-01 -1.63874530e-01 2.82887658e-01 -1.84342587e-01 -2.06414411e-01 -2.08204480e-01 -1.38729793e-02 -1.57740448e-01 -2.15503443e-01 -2.28555304e-02 1.51489503e-02 7.84429749e-02 2.26152501e-02 6.08806041e-02 7.26464614e-02 5.50632050e-02 6.73235407e-02 6.35158096e-02 -1.68722251e-01 6.04495891e-02 2.65592982e-02 -2.80956494e-02 -4.57351577e-02 3.68609334e-02 -2.53026387e-02 -1.07560260e-01 2.16413638e-01 2.64175182e-01 2.61845376e-01 2.20263366e-01 5.33270931e-02 1.16306645e-01 -2.54853144e-01 -1.85091925e-01 2.05276019e-02 -1.22967852e-02 -1.23565806e-01 -1.88534745e-01 4.23969750e-02]]
X_set, y_set = x_test_pca, y_test
X1, X2 = np.meshgrid(np.arange(start = X_set[:, 0].min() - 1,
stop = X_set[:, 0].max() + 1,
step = 0.01),
np.arange(start = X_set[:, 1].min() - 1,
stop = X_set[:, 1].max() + 1,
step = 0.01))
#plt.contourf(X1,X2,
# lr.predict(np.array([X1.ravel(),X2.ravel()]).T).reshape(X1.shape),
# alpha = 0.75, cmap = ListedColormap(('red', 'green', 'blue')))
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
plt.scatter(X_set[y_set == j, 0], X_set[y_set == j, 2],
c = ListedColormap(('darkorange', 'teal', 'violet'))(i), label = j)
plt.title('Logistic Regression (Test set)')
plt.xlabel('PCA1')
plt.ylabel('PCA3')
plt.legend()
#plt.savefig("/content/drive/My Drive/wp1/scatter13_pca1.png", dpi=300)
plt.show()
WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points.
speakers = set(data['participant'].unique().tolist())
#speakers = {{'P0001', 'P0002', 'P0003', 'P0004'}}
print(speakers)
train_scores = []
test_scores=[]
f1 = []
Y_tests = []
Y_preds= []
PCA_var = []
for sp in tqdm(speakers):
# define training and test sets
print(sp)
train_sp = list(speakers.difference([sp]))
train_data = data[data['participant'].isin(train_sp)]
test_data = data[data['participant'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:62]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:62]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
# Apply PCA
pca = PCA(n_components=5)
x_train_pca = pca.fit_transform(x_train_scaled)
# Initialize the logistic regression model
# Train the model
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=5000)
lr.fit(x_train_pca,y_train)
# Measure accuracy
x_test = scaler.transform(x_test)
x_test_pca = pca.transform(x_test)
test_scores.append(lr.score(x_test_pca, y_test))
train_scores.append(accuracy_score(y_train, lr.predict(x_train_pca)))
y_pred = lr.predict(x_test_pca)
f1.append(f1_score(y_test, y_pred, average='macro'))
PCA_var.append(pca.explained_variance_ratio_.sum())
Y_tests.extend(y_test)
Y_preds.extend(y_pred)
idx = np.argsort(lr.coef_[0])
{'P0003', 'P0001', 'P0002', 'P0004'}
0%| | 0/4 [00:00<?, ?it/s]
P0003
25%|██▌ | 1/4 [00:00<00:01, 2.82it/s]
P0001
50%|█████ | 2/4 [00:00<00:00, 3.36it/s]
P0002
75%|███████▌ | 3/4 [00:01<00:00, 2.71it/s]
P0004
100%|██████████| 4/4 [00:01<00:00, 3.00it/s]
# report accuracy
print("Train accuracy:", np.round(np.mean(train_scores), 2))
print("Test accuracy:", np.round(np.mean(test_scores), 2))
print(np.mean(f1))
print(PCA_var)
print(np.round(np.mean(PCA_var),2))
# Make the confusion matrix (aggregated)
from sklearn.metrics import confusion_matrix
cf_matrix = confusion_matrix(Y_tests, Y_preds)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Train accuracy: 0.92 Test accuracy: 0.78 0.775309412928585 [0.6305459514935886, 0.621689962855177, 0.6435861932901861, 0.6090960926261011] 0.63 Test confusion_matrix
Text(33.0, 0.5, 'True')
# Make the confusion matrix (last iteration)
cf_matrix = confusion_matrix(y_test, y_pred)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Test confusion_matrix
Text(33.0, 0.5, 'True')
test_scores
[0.7839506172839507, 0.7716049382716049, 0.7530864197530864, 0.8148148148148148]
# The coefficients
print('Coefficients: \n', lr.coef_)
fig = plt.figure(figsize=(25, 10))
ax = sns.heatmap(pca.components_,
cmap='PiYG',
yticklabels=[ "PCA"+str(x) for x in range(1,16)],
xticklabels=list(x_train.columns))
#cbar_kws={"orientation": "horizontal"})
#ax.set_aspect("equal")
#fig.savefig("/content/drive/My Drive/wp1/heatmap_pca2.png", dpi= 300)
Coefficients: [[-0.96838678 -0.57774519 1.42527163 0. 0.89628569] [ 2.51941865 0. -2.85986632 0.92601847 -1.82061765] [ 0. 0.29193453 0. -0.12713881 0. ]]
# The coefficients
print('Coefficients: \n', lr.coef_)
fig = plt.figure(figsize=(10, 30))
ax = sns.heatmap(pca.components_.T, annot=True,
cmap='PiYG',
norm=colors.Normalize(),
xticklabels=[ "PCA"+str(x) for x in range(1,6)],
yticklabels=list(x_train.columns),
cbar_kws={"orientation": "horizontal"})
#ax.set_aspect("1")
#fig.savefig("/content/drive/My Drive/wp1/heatmap_pca2shortv.png", dpi=300)
Coefficients: [[-0.96838678 -0.57774519 1.42527163 0. 0.89628569] [ 2.51941865 0. -2.85986632 0.92601847 -1.82061765] [ 0. 0.29193453 0. -0.12713881 0. ]]
X_set, y_set = x_test_pca, y_test
X1, X2 = np.meshgrid(np.arange(start = X_set[:, 0].min() - 1,
stop = X_set[:, 0].max() + 1,
step = 0.01),
np.arange(start = X_set[:, 1].min() - 1,
stop = X_set[:, 1].max() + 1,
step = 0.01))
#plt.contourf(X1,X2,
# lr.predict(np.array([X1.ravel(),X2.ravel()]).T).reshape(X1.shape),
# alpha = 0.75, cmap = ListedColormap(('red', 'green', 'blue')))
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
plt.scatter(X_set[y_set == j, 0], X_set[y_set == j, 2],
c = ListedColormap(('darkorange', 'teal', 'violet'))(i), label = j)
plt.title('Logistic Regression (Test set)')
plt.xlabel('PC1')
plt.ylabel('PC3')
plt.legend()
#plt.savefig("/content/drive/My Drive/wp1/scatter13_pca2.png", dpi=300)
plt.show()
WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points.
def biplot(score,coeff,labels=None):
xs = score[:,0]
ys = score[:,1]
n = coeff.shape[0]
scalex = 1.0/(xs.max() - xs.min())
scaley = 1.0/(ys.max() - ys.min())
plt.scatter(xs * scalex,ys * scaley,s=5)
for i in range(n):
plt.arrow(0, 0, coeff[i,0], coeff[i,1],color = 'r',alpha = 0.5)
if labels is None:
plt.text(coeff[i,0]* 1.15, coeff[i,1] * 1.15, "Var"+str(i+1), color = 'green', ha = 'center', va = 'center')
else:
plt.text(coeff[i,0]* 1.15, coeff[i,1] * 1.15, labels[i], color = 'g', ha = 'center', va = 'center')
plt.xlabel("PC{}".format(1))
plt.ylabel("PC{}".format(2))
plt.grid()
biplot(x_train_pca[:,0:2], np.transpose(pca.components_[0:2,:]), list(x_train.columns))
plt.show()
#from matplotlib.colors import ListedColormap
X_set, y_set = x_test_pca, y_test
X1, X2 = np.meshgrid(np.arange(start = X_set[:, 0].min() - 1,
stop = X_set[:, 0].max() + 1,
step = 0.01),
np.arange(start = X_set[:, 1].min() - 1,
stop = X_set[:, 1].max() + 1,
step = 0.01))
#plt.contourf(X1,X2,
# lr.predict(np.array([X1.ravel(),X2.ravel()]).T).reshape(X1.shape),
# alpha = 0.75, cmap = ListedColormap(('red', 'green', 'blue')))
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
plt.scatter(X_set[y_set == j, 0], X_set[y_set == j, 1],
c = ListedColormap(('darkorange', 'teal', 'violet'))(i), label = j)
plt.title('Logistic Regression (Test set)')
plt.xlabel('PCA3')
plt.ylabel('PCA5')
plt.legend()
#plt.savefig("/content/drive/My Drive/wp1/scatter34_pca2.png", dpi=300)
plt.show()
WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points.
#from matplotlib.colors import ListedColormap
X_set, y_set = x_test_pca, y_test
X1, X2 = np.meshgrid(np.arange(start = X_set[:, 0].min() - 1,
stop = X_set[:, 0].max() + 1,
step = 0.01),
np.arange(start = X_set[:, 1].min() - 1,
stop = X_set[:, 1].max() + 1,
step = 0.01))
#plt.contourf(X1,X2,
# lr.predict(np.array([X1.ravel(),X2.ravel()]).T).reshape(X1.shape),
# alpha = 0.75, cmap = ListedColormap(('red', 'green', 'blue')))
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
plt.scatter(X_set[y_set == j, 0], X_set[y_set == j, 1],
c = ListedColormap(('darkorange', 'teal', 'violet'))(i), label = j)
plt.title('Logistic Regression (Test set)')
plt.xlabel('PCA1')
plt.ylabel('PCA3')
plt.legend()
#plt.savefig("/content/drive/My Drive/wp1/scatter13_pca2.png", dpi=300)
plt.show()
WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points.
Both PCA and LDA are linear transformation techniques. However, PCA is an unsupervised while LDA is a supervised dimensionality reduction technique.
PCA has no concern with the class labels. In simple words, PCA summarizes the feature set without relying on the output. PCA tries to find the directions of the maximum variance in the dataset.
LDA tries to reduce dimensions of the feature set while retaining the information that discriminates output classes.
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
sessions = set(data['rs'].unique().tolist())
test_scores_lda = []
test_scores_lr =[]
f1 = []
Y_tests = []
Y_preds= []
eigen = []
for sp in tqdm(sessions):
# define training and test sets
train_sp = list(sessions.difference([sp]))
train_data = data[data['rs'].isin(train_sp)]
test_data = data[data['rs'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:64]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:64]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
x_test = scaler.transform(x_test)
# Apply LDA
lda = LinearDiscriminantAnalysis()
x_train_lda = lda.fit_transform(x_train_scaled, y_train)
x_test_lda = lda.transform(x_test)
# Initialize the logistic regression classifier
# Train the classifier model
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=5000)
lr.fit(x_train_lda,y_train)
# Measure accuracy
y_pred1 = lda.predict(x_test) #ida prediction
y_pred2 = lr.predict(x_test_lda) #lr prediction
test_scores_lda.append(accuracy_score(y_pred1, y_test))
test_scores_lr.append(accuracy_score(y_pred2, y_test))
f1.append(f1_score(y_test, y_pred2, average='macro'))
Y_tests.extend(y_test)
Y_preds.extend(y_pred2)
idx = np.argsort(lda.coef_[0])
100%|██████████| 9/9 [00:02<00:00, 3.28it/s]
# report accuracy
print("Test accuracy (LDA):", np.round(np.mean(test_scores_lda), 2))
print("Test accuracy (LDA + LR):", np.round(np.mean(test_scores_lr), 2))
print(np.mean(f1))
print(test_scores_lda)
print(test_scores_lr)
print(lda.explained_variance_ratio_.sum())
# Make the confusion matrix (aggregated)
from sklearn.metrics import confusion_matrix
cf_matrix = confusion_matrix(Y_tests, Y_preds)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Test accuracy (LDA): 0.98 Test accuracy (LDA + LR): 0.98 0.9814379188479447 [0.9861111111111112, 1.0, 1.0, 0.9861111111111112, 1.0, 0.9722222222222222, 0.9861111111111112, 0.9722222222222222, 0.9583333333333334] [0.9861111111111112, 0.9861111111111112, 0.9722222222222222, 0.9861111111111112, 0.9861111111111112, 0.9861111111111112, 0.9861111111111112, 0.9861111111111112, 0.9583333333333334] 1.0 Test confusion_matrix
Text(33.0, 0.5, 'True')
from matplotlib.colors import ListedColormap
X_set, y_set = x_test_lda, y_test
X1, X2 = np.meshgrid(np.arange(start = X_set[:, 0].min() - 1,
stop = X_set[:, 0].max() + 1,
step = 0.01),
np.arange(start = X_set[:, 1].min() - 1,
stop = X_set[:, 1].max() + 1,
step = 0.01))
plt.contourf(X1,X2,
lr.predict(np.array([X1.ravel(),X2.ravel()]).T).reshape(X1.shape),
alpha = 0.5, cmap = ListedColormap(('darkorange', 'teal', 'violet')))
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
plt.scatter(X_set[y_set == j, 0], X_set[y_set == j, 1],
c = ListedColormap(( 'darkorange', 'teal', 'violet'))(i), label = j)
plt.title('Logistic Regression (Test set)')
plt.xlabel('LD1')
plt.ylabel('LD2')
plt.legend()
#plt.savefig("/content/drive/My Drive/wp1/lda1_test.png", dpi=300)
plt.show()
WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points.
# The coefficients
print('Coefficients: \n', lr.coef_)
fig = plt.figure(figsize=(20, 10))
ax = sns.heatmap(lda.scalings_.T,
cmap='PiYG',
norm=colors.Normalize(vmin=-220.,vmax=220.),
yticklabels=[ "LDA"+str(x) for x in range(1,3)],
xticklabels=list(x_train.columns))
#cbar_kws={"orientation": "horizontal"})
#ax.set_aspect("equal")
#fig.savefig("/content/drive/My Drive/wp1/heatmap_lda1.png", dpi=300)
Coefficients: [[-3.03245356 -0.02941438] [ 6.07691933 0. ] [ 0. 2.64675245]]
Both PCA and LDA are linear transformation techniques. However, PCA is an unsupervised while LDA is a supervised dimensionality reduction technique.
PCA has no concern with the class labels. In simple words, PCA summarizes the feature set without relying on the output. PCA tries to find the directions of the maximum variance in the dataset.
LDA tries to reduce dimensions of the feature set while retaining the information that discriminates output classes.
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
speakers = set(data['participant'].unique().tolist())
test_scores_lda = []
test_scores_lr=[]
f1 = []
Y_tests = []
Y_preds= []
eigen = []
for sp in tqdm(speakers):
# define training and test sets
train_sp = list(speakers.difference([sp]))
train_data = data[data['participant'].isin(train_sp)]
test_data = data[data['participant'] == sp]
# Encoding of labels
y_train = train_data['phonation'].tolist()
le = LabelEncoder()
y_train = le.fit_transform(y_train)
x_train = train_data.iloc[:,4:64]
y_test = test_data['phonation'].tolist()
y_test = le.transform(y_test)
x_test = test_data.iloc[:,4:64]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
x_test = scaler.transform(x_test)
# Apply LDA
lda = LinearDiscriminantAnalysis()
x_train_lda = lda.fit_transform(x_train_scaled, y_train)
x_test_lda = lda.transform(x_test)
# Initialize the logistic regression model
# Train the classifier model
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=5000)
lr.fit(x_train_lda,y_train)
# Measure accuracy
y_pred1 = lda.predict(x_test)
y_pred2 = lr.predict(x_test_lda)
test_scores_lda.append(accuracy_score(y_pred1, y_test))
test_scores_lr.append(accuracy_score(y_pred2, y_test))
f1.append(f1_score(y_test, y_pred2, average='macro'))
Y_tests.extend(y_test)
Y_preds.extend(y_pred2)
idx = np.argsort(lda.coef_[0])
100%|██████████| 4/4 [00:00<00:00, 4.39it/s]
# report accuracy
print("Test accuracy (LDA):", np.round(np.mean(test_scores_lda), 2))
print("Test accuracy (LDA + LR):", np.round(np.mean(test_scores_lr), 2))
print(np.mean(f1))
print(test_scores_lda)
print(test_scores_lr)
print(lda.explained_variance_ratio_.sum())
# Make the confusion matrix (aggregated)
from sklearn.metrics import confusion_matrix
cf_matrix = confusion_matrix(Y_tests, Y_preds)
print("\nTest confusion_matrix")
sns.heatmap(cf_matrix, annot=True, cmap='Blues')
plt.xlabel('Predicted', fontsize=12)
plt.ylabel('True', fontsize=12)
Test accuracy (LDA): 0.84 Test accuracy (LDA + LR): 0.79 0.7701850317969695 [0.7345679012345679, 0.9259259259259259, 0.7901234567901234, 0.9259259259259259] [0.7222222222222222, 0.8024691358024691, 0.7654320987654321, 0.8518518518518519] 1.0 Test confusion_matrix
Text(33.0, 0.5, 'True')
from matplotlib.colors import ListedColormap
X_set, y_set = x_test_lda, y_test
X1, X2 = np.meshgrid(np.arange(start = X_set[:, 0].min() - 1,
stop = X_set[:, 0].max() + 1,
step = 0.01),
np.arange(start = X_set[:, 1].min() - 1,
stop = X_set[:, 1].max() + 1,
step = 0.01))
plt.contourf(X1,X2,
lr.predict(np.array([X1.ravel(),X2.ravel()]).T).reshape(X1.shape),
alpha = 0.5, cmap = ListedColormap(('darkorange', 'teal', 'violet')))
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
plt.scatter(X_set[y_set == j, 0], X_set[y_set == j, 1],
c = ListedColormap(( 'darkorange', 'teal', 'violet'))(i), label = j)
plt.title('Logistic Regression (Test set)')
plt.xlabel('LD1')
plt.ylabel('LD2')
plt.legend()
#plt.savefig("/content/drive/My Drive/wp1/lda2_test.png", dpi=300)
plt.show()
WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points. WARNING:matplotlib.axes._axes:*c* argument looks like a single numeric RGB or RGBA sequence, which should be avoided as value-mapping will have precedence in case its length matches with *x* & *y*. Please use the *color* keyword-argument or provide a 2-D array with a single row if you intend to specify the same RGB or RGBA value for all points.
# The coefficients
print('Coefficients: \n', lr.coef_)
import matplotlib.colors as colors
fig = plt.figure(figsize=(20, 10))
ax = sns.heatmap(lda.scalings_.T,
cmap='PiYG',
norm=colors.Normalize(vmin=-220.,vmax=220.),
yticklabels=[ "LDA"+str(x) for x in range(1,3)],
xticklabels=list(x_train.columns))
#cbar_kws={"orientation": "horizontal"})
#ax.set_aspect("equal")
fig.savefig("/content/drive/My Drive/wp1/heatmap_lda2.png", dpi=300)
Coefficients: [[-4.18400492 -2.41634433] [ 3.09693628 0. ] [ 0. 4.52132167]]
lda.scalings_
array([[-6.38903532e+00, -5.74817738e+01],
[ 8.16986209e+01, -7.59508767e+01],
[-1.16980440e+02, -1.50443460e+02],
[ 1.42738500e+01, 1.02671113e+02],
[ 6.80187844e+01, 1.17999397e+02],
[-6.67730216e+01, 7.46152096e+01],
[ 3.28068533e+01, 1.62258075e+01],
[-4.52792573e+00, -2.21145209e+02],
[-3.57406425e+01, -5.63399764e+01],
[ 1.42016704e+01, 1.13040291e+02],
[ 6.32486461e+01, 1.07960632e+02],
[-8.23976529e+01, 9.27596191e+01],
[ 2.66654818e+01, 1.27904660e+01],
[ 3.63445893e-01, -3.66815409e-01],
[-5.64397442e-01, 1.15598749e+00],
[-1.45485058e-01, 5.30038880e-01],
[-5.38683676e-01, -2.17513381e-01],
[ 4.83038353e-01, 1.16846588e+00],
[-1.70462664e-01, -6.61528284e-01],
[-7.70860712e-01, 5.12148829e-01],
[-8.23091139e-01, 1.03887749e+00],
[ 3.53752998e-01, -1.27353183e-01],
[-1.90150730e-01, -5.53468487e-01],
[ 2.09476420e-01, -4.43628882e-01],
[ 4.03185763e-01, 2.11449385e-01],
[-9.18916512e-01, -1.51746116e+00],
[-1.61347352e-01, -1.49418299e-02],
[-1.89380708e-01, 3.84227788e-01],
[-2.85501862e-01, -6.34410362e-02],
[ 4.40719976e-03, 6.99933246e-01],
[-1.77419864e-01, -5.07989394e-02],
[ 2.80862289e-01, 5.64695572e-01],
[-1.99749046e-01, -3.77839595e-01],
[-5.71346760e-02, 2.21746617e-01],
[ 1.15533694e-01, -5.49185265e-01],
[ 3.44191119e-01, 3.17873324e-01],
[ 7.95740657e-02, -5.76238990e-01],
[-2.11899211e-03, -3.28334481e-01],
[-2.33944431e-01, -1.78880906e-01],
[-1.02094864e-01, -1.38695026e-02],
[ 2.54492492e-02, 9.19916769e-02],
[ 7.41452575e-02, -4.61636387e-01],
[-2.48107683e-02, 2.24075759e-01],
[-2.58967456e-01, 8.63575205e-02],
[ 2.26281125e-01, -6.31291600e-01],
[ 3.17632543e-01, -3.36733738e-01],
[-5.53403156e-02, 1.96975980e-01],
[ 7.23011230e-02, 1.82943110e-01],
[ 5.59579727e-01, -5.88326830e-02],
[-5.60198406e-01, -3.21674942e-01],
[ 3.58280899e-01, 2.54785432e-01],
[-2.36620661e-01, 1.41970802e-01],
[ 3.48538450e-02, 2.21865500e-01],
[-9.79214791e-02, 1.37270275e-01],
[ 1.64551962e-01, 1.84808234e-01],
[ 2.70581811e-01, 4.08279880e-01],
[-4.85742937e-02, 1.18373034e-01],
[ 6.47525682e-02, -1.40575581e-01],
[ 2.61619598e-02, 2.27536601e-02],
[-5.57947444e-01, -1.32864093e-01]])
# Some biplot and scatter plot attempts
# def bildaplot(score,coeff,labels=None):
# xs = score[:,0]
# ys = score[:,1]
# n = coeff.shape[0]
# plt.scatter(xs, ys, c=y_test)
# for i in range(n):
# plt.arrow(0, 0, coeff[i,0], coeff[i,1],color = 'r',alpha = 0.5)
# if labels is None:
# plt.text(coeff[i,0]* 1.15, coeff[i,1] * 1.15, "Var"+str(i+1),
# color = 'g', ha = 'center', va = 'center')
# else:
# plt.text(coeff[i,0]* 1.15, coeff[i,1] * 1.15, labels[i],
# color = 'g', ha = 'center', va = 'center')
# xs = x_test_lda[:,0]
# ys = x_test_lda[:,1]
# plt.scatter(xs, ys, c=y_test)
# plt.xlabel("LD{}".format(1))
# plt.ylabel("LD{}".format(2))
# plt.grid()
# plt.legend()
# #Call the function.
# # bildaplot(x_test_lda[:,0:2], lda.scalings_) #scales feel wrong.
# plt.savefig("/content/drive/My Drive/wp1/lda2.png", dpi=300)
# plt.show()
WARNING:matplotlib.legend:No handles with labels found to put in legend.
# train and test set
x = data.iloc[:,4:64]
scaler = StandardScaler()
X = scaler.fit_transform(x)
y = data['phonation'].tolist()
le = LabelEncoder()
y = le.fit_transform(y)
target_names = le.classes_
pca = PCA(n_components=2)
X_r = pca.fit(X).transform(X)
lda = LinearDiscriminantAnalysis(n_components=2)
X_r2 = lda.fit(X, y).transform(X)
# Percentage of variance explained for each components
print('explained variance ratio (first two components): %s'
% str(pca.explained_variance_ratio_))
print('explained variance ratio (first two components): %s'
% str(lda.explained_variance_ratio_))
plt.figure()
for c, i, target_name in zip("rgb", [0, 1, 2], target_names):
plt.scatter(X_r[y == i, 0], X_r[y == i, 1], c=c, label=target_name)
plt.legend()
plt.title('PCA')
plt.figure()
for c, i, target_name in zip("rgb", [0, 1, 2], target_names):
plt.scatter(X_r2[y == i, 0], X_r2[y == i, 1], c=c, label=target_name)
plt.legend()
plt.title('LDA')
plt.show()
explained variance ratio (first two components): [0.20262207 0.14816274] explained variance ratio (first two components): [0.84205883 0.15794117]
Combining VoiceSauce and Opensmile data
dvs = pd.read_csv('/content/drive/My Drive/wp1/vsdatas.csv', na_filter = False)
dos = pd.read_csv('/content/drive/My Drive/wp1/opensmile1-4funcs.csv', na_filter = False)
data = pd.merge(dvs, dos, on=['participant','session','repetition','phonation','comb_no'],
left_index=False, right_index=False,
how="outer")
data
| participant | Filename | phonation | H1c_mean | H2c_mean | H4c_mean | A1c_mean | A2c_mean | A3c_mean | H2Kc_mean | ... | slopeUV0-500_sma3nz_amean | slopeUV500-1500_sma3nz_amean | spectralFluxUV_sma3nz_amean | loudnessPeaksPerSec | VoicedSegmentsPerSec | MeanVoicedSegmentLengthSec | StddevVoicedSegmentLengthSec | MeanUnvoicedSegmentLength | StddevUnvoicedSegmentLength | equivalentSoundLevel_dBp | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | P0001 | AC_P0001_S01_R01_BRT_HDM.mat | BRT | 16.425432 | 5.149999 | -1.754374 | -15.304558 | -14.625361 | -5.760611 | -3.765104 | ... | -0.055642 | 0.004372 | 0.082026 | 4.228687 | 2.304304 | 0.248676 | 0.202455 | 0.171538 | 0.251367 | -38.744747 |
| 1 | P0001 | AC_P0001_S01_R01_CRK_HDM.mat | CRK | 13.142466 | 10.741680 | 3.306533 | -9.268425 | -15.097425 | -6.427032 | -2.980318 | ... | -0.053462 | 0.000241 | 0.087053 | 4.819720 | 2.359012 | 0.276406 | 0.255706 | 0.128065 | 0.180544 | -41.022310 |
| 2 | P0001 | AC_P0001_S01_R01_MOD_HDM.mat | MOD | 13.814134 | 7.490530 | 0.251819 | -9.497376 | -14.923896 | -5.244940 | -2.523498 | ... | -0.059919 | 0.002522 | 0.085262 | 4.662439 | 2.017937 | 0.320370 | 0.248372 | 0.150364 | 0.215600 | -37.306293 |
| 3 | P0001 | AC_P0001_S01_R02_BRT_HDM.mat | BRT | 14.824055 | 5.192939 | -0.332665 | -11.515982 | -11.647516 | -3.783325 | -2.726151 | ... | -0.042710 | -0.000428 | 0.091031 | 3.932773 | 2.356902 | 0.237143 | 0.203429 | 0.167681 | 0.222614 | -36.364433 |
| 4 | P0001 | AC_P0001_S01_R02_CRK_HDM.mat | CRK | 14.185648 | 12.096453 | 5.616025 | -8.075565 | -15.252716 | -7.001800 | -3.781944 | ... | -0.055427 | -0.002846 | 0.094226 | 4.783841 | 2.094427 | 0.311186 | 0.293547 | 0.163462 | 0.249252 | -39.633247 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 112 | P0004 | AC_P0004_S03_R02_CRK_HDM.mat | CRK | 14.403799 | 11.561428 | 4.453420 | -6.904047 | -14.080556 | -8.114757 | -4.335758 | ... | -0.067100 | 0.000034 | 0.066857 | 3.299024 | 1.427061 | 0.487963 | 0.414873 | 0.215208 | 0.324859 | -36.072060 |
| 113 | P0004 | AC_P0004_S03_R02_MOD_HDM.mat | MOD | 16.753962 | 11.389900 | 4.531010 | -6.086536 | -11.474333 | -4.162062 | -0.862434 | ... | -0.054319 | 0.000623 | 0.073302 | 3.504604 | 1.516954 | 0.467059 | 0.416975 | 0.173400 | 0.266170 | -34.118713 |
| 114 | P0004 | AC_P0004_S03_R03_BRT_HDM.mat | BRT | 17.605715 | 6.205063 | -0.576680 | -14.470706 | -17.842321 | -9.860130 | -7.101625 | ... | -0.057897 | 0.001670 | 0.053314 | 3.710938 | 1.536742 | 0.422727 | 0.373083 | 0.219423 | 0.329425 | -37.928776 |
| 115 | P0004 | AC_P0004_S03_R03_CRK_HDM.mat | CRK | 12.654315 | 9.636426 | 2.572714 | -9.056556 | -17.520300 | -9.791619 | -6.583420 | ... | -0.067164 | -0.001297 | 0.060492 | 3.163152 | 1.500417 | 0.481296 | 0.399569 | 0.171961 | 0.250192 | -37.965910 |
| 116 | P0004 | AC_P0004_S03_R03_MOD_HDM.mat | MOD | 15.900629 | 10.546235 | 2.681915 | -8.103086 | -13.164056 | -6.177027 | -1.827386 | ... | -0.063104 | 0.002309 | 0.068834 | 3.375404 | 1.824066 | 0.371935 | 0.310945 | 0.160167 | 0.251485 | -35.056450 |
117 rows × 157 columns
# Rearrange the columns
data.insert(1, 'file', data.pop('file'))
data.insert(4, 'start', data.pop('start'))
data.insert(5, 'end', data.pop('end'))
data.insert(1, 'session', data.pop('session'))
data.insert(2, 'repetition', data.pop('repetition'))
data.insert(3, 'comb_no', data.pop('comb_no'))
data.insert(4, 'phonation', data.pop('phonation'))
data
| participant | session | repetition | comb_no | phonation | file | Filename | start | end | H1c_mean | ... | slopeUV0-500_sma3nz_amean | slopeUV500-1500_sma3nz_amean | spectralFluxUV_sma3nz_amean | loudnessPeaksPerSec | VoicedSegmentsPerSec | MeanVoicedSegmentLengthSec | StddevVoicedSegmentLengthSec | MeanUnvoicedSegmentLength | StddevUnvoicedSegmentLength | equivalentSoundLevel_dBp | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | P0001 | S01 | R01 | S01R01 | BRT | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R01_BRT_HDM.mat | 0 days | 0 days 00:00:29.577208333 | 16.425432 | ... | -0.055642 | 0.004372 | 0.082026 | 4.228687 | 2.304304 | 0.248676 | 0.202455 | 0.171538 | 0.251367 | -38.744747 |
| 1 | P0001 | S01 | R01 | S01R01 | CRK | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R01_CRK_HDM.mat | 0 days | 0 days 00:00:27.197750 | 13.142466 | ... | -0.053462 | 0.000241 | 0.087053 | 4.819720 | 2.359012 | 0.276406 | 0.255706 | 0.128065 | 0.180544 | -41.022310 |
| 2 | P0001 | S01 | R01 | S01R01 | MOD | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R01_MOD_HDM.mat | 0 days | 0 days 00:00:26.828604167 | 13.814134 | ... | -0.059919 | 0.002522 | 0.085262 | 4.662439 | 2.017937 | 0.320370 | 0.248372 | 0.150364 | 0.215600 | -37.306293 |
| 3 | P0001 | S01 | R02 | S01R02 | BRT | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R02_BRT_HDM.mat | 0 days | 0 days 00:00:29.761750 | 14.824055 | ... | -0.042710 | -0.000428 | 0.091031 | 3.932773 | 2.356902 | 0.237143 | 0.203429 | 0.167681 | 0.222614 | -36.364433 |
| 4 | P0001 | S01 | R02 | S01R02 | CRK | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R02_CRK_HDM.mat | 0 days | 0 days 00:00:28.230791667 | 14.185648 | ... | -0.055427 | -0.002846 | 0.094226 | 4.783841 | 2.094427 | 0.311186 | 0.293547 | 0.163462 | 0.249252 | -39.633247 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 112 | P0004 | S03 | R02 | S03R02 | CRK | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R02_CRK_HDM.mat | 0 days | 0 days 00:00:37.903666667 | 14.403799 | ... | -0.067100 | 0.000034 | 0.066857 | 3.299024 | 1.427061 | 0.487963 | 0.414873 | 0.215208 | 0.324859 | -36.072060 |
| 113 | P0004 | S03 | R02 | S03R02 | MOD | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R02_MOD_HDM.mat | 0 days | 0 days 00:00:33.685812500 | 16.753962 | ... | -0.054319 | 0.000623 | 0.073302 | 3.504604 | 1.516954 | 0.467059 | 0.416975 | 0.173400 | 0.266170 | -34.118713 |
| 114 | P0004 | S03 | R03 | S03R03 | BRT | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R03_BRT_HDM.mat | 0 days | 0 days 00:00:35.859729167 | 17.605715 | ... | -0.057897 | 0.001670 | 0.053314 | 3.710938 | 1.536742 | 0.422727 | 0.373083 | 0.219423 | 0.329425 | -37.928776 |
| 115 | P0004 | S03 | R03 | S03R03 | CRK | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R03_CRK_HDM.mat | 0 days | 0 days 00:00:36.052770833 | 12.654315 | ... | -0.067164 | -0.001297 | 0.060492 | 3.163152 | 1.500417 | 0.481296 | 0.399569 | 0.171961 | 0.250192 | -37.965910 |
| 116 | P0004 | S03 | R03 | S03R03 | MOD | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R03_MOD_HDM.mat | 0 days | 0 days 00:00:34.081312500 | 15.900629 | ... | -0.063104 | 0.002309 | 0.068834 | 3.375404 | 1.824066 | 0.371935 | 0.310945 | 0.160167 | 0.251485 | -35.056450 |
117 rows × 157 columns
# average duration of audios
dur = data['end'].str.replace("0 days 00:00:","")
dur = pd.to_numeric(dur)
print(dur.mean())
print(dur.std(ddof=1))
29.040706908803415 3.7739142851002523
# See all column names
print(data.columns.tolist())
['participant', 'session', 'repetition', 'comb_no', 'phonation', 'file', 'Filename', 'start', 'end', 'H1c_mean', 'H2c_mean', 'H4c_mean', 'A1c_mean', 'A2c_mean', 'A3c_mean', 'H2Kc_mean', 'H1H2c_mean', 'H2H4c_mean', 'H1A1c_mean', 'H1A2c_mean', 'H1A3c_mean', 'H42Kc_mean', 'H2KH5Kc_mean', 'CPP_mean', 'Energy_mean', 'HNR05_mean', 'HNR15_mean', 'HNR25_mean', 'HNR35_mean', 'sF0_mean', 'sF1_mean', 'sF2_mean', 'sF3_mean', 'sF4_mean', 'sB1_mean', 'sB2_mean', 'sB3_mean', 'sB4_mean', 'H1c_sd', 'H2c_sd', 'H4c_sd', 'A1c_sd', 'A2c_sd', 'A3c_sd', 'H2Kc_sd', 'H1H2c_sd', 'H2H4c_sd', 'H1A1c_sd', 'H1A2c_sd', 'H1A3c_sd', 'H42Kc_sd', 'H2KH5Kc_sd', 'CPP_sd', 'Energy_sd', 'HNR05_sd', 'HNR15_sd', 'HNR25_sd', 'HNR35_sd', 'sF0_sd', 'sF1_sd', 'sF2_sd', 'sF3_sd', 'sF4_sd', 'sB1_sd', 'sB2_sd', 'sB3_sd', 'sB4_sd', 'F0semitoneFrom27.5Hz_sma3nz_amean', 'F0semitoneFrom27.5Hz_sma3nz_stddevNorm', 'F0semitoneFrom27.5Hz_sma3nz_percentile20.0', 'F0semitoneFrom27.5Hz_sma3nz_percentile50.0', 'F0semitoneFrom27.5Hz_sma3nz_percentile80.0', 'F0semitoneFrom27.5Hz_sma3nz_pctlrange0-2', 'F0semitoneFrom27.5Hz_sma3nz_meanRisingSlope', 'F0semitoneFrom27.5Hz_sma3nz_stddevRisingSlope', 'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope', 'F0semitoneFrom27.5Hz_sma3nz_stddevFallingSlope', 'loudness_sma3_amean', 'loudness_sma3_stddevNorm', 'loudness_sma3_percentile20.0', 'loudness_sma3_percentile50.0', 'loudness_sma3_percentile80.0', 'loudness_sma3_pctlrange0-2', 'loudness_sma3_meanRisingSlope', 'loudness_sma3_stddevRisingSlope', 'loudness_sma3_meanFallingSlope', 'loudness_sma3_stddevFallingSlope', 'spectralFlux_sma3_amean', 'spectralFlux_sma3_stddevNorm', 'mfcc1_sma3_amean', 'mfcc1_sma3_stddevNorm', 'mfcc2_sma3_amean', 'mfcc2_sma3_stddevNorm', 'mfcc3_sma3_amean', 'mfcc3_sma3_stddevNorm', 'mfcc4_sma3_amean', 'mfcc4_sma3_stddevNorm', 'jitterLocal_sma3nz_amean', 'jitterLocal_sma3nz_stddevNorm', 'shimmerLocaldB_sma3nz_amean', 'shimmerLocaldB_sma3nz_stddevNorm', 'HNRdBACF_sma3nz_amean', 'HNRdBACF_sma3nz_stddevNorm', 'logRelF0-H1-H2_sma3nz_amean', 'logRelF0-H1-H2_sma3nz_stddevNorm', 'logRelF0-H1-A3_sma3nz_amean', 'logRelF0-H1-A3_sma3nz_stddevNorm', 'F1frequency_sma3nz_amean', 'F1frequency_sma3nz_stddevNorm', 'F1bandwidth_sma3nz_amean', 'F1bandwidth_sma3nz_stddevNorm', 'F1amplitudeLogRelF0_sma3nz_amean', 'F1amplitudeLogRelF0_sma3nz_stddevNorm', 'F2frequency_sma3nz_amean', 'F2frequency_sma3nz_stddevNorm', 'F2bandwidth_sma3nz_amean', 'F2bandwidth_sma3nz_stddevNorm', 'F2amplitudeLogRelF0_sma3nz_amean', 'F2amplitudeLogRelF0_sma3nz_stddevNorm', 'F3frequency_sma3nz_amean', 'F3frequency_sma3nz_stddevNorm', 'F3bandwidth_sma3nz_amean', 'F3bandwidth_sma3nz_stddevNorm', 'F3amplitudeLogRelF0_sma3nz_amean', 'F3amplitudeLogRelF0_sma3nz_stddevNorm', 'alphaRatioV_sma3nz_amean', 'alphaRatioV_sma3nz_stddevNorm', 'hammarbergIndexV_sma3nz_amean', 'hammarbergIndexV_sma3nz_stddevNorm', 'slopeV0-500_sma3nz_amean', 'slopeV0-500_sma3nz_stddevNorm', 'slopeV500-1500_sma3nz_amean', 'slopeV500-1500_sma3nz_stddevNorm', 'spectralFluxV_sma3nz_amean', 'spectralFluxV_sma3nz_stddevNorm', 'mfcc1V_sma3nz_amean', 'mfcc1V_sma3nz_stddevNorm', 'mfcc2V_sma3nz_amean', 'mfcc2V_sma3nz_stddevNorm', 'mfcc3V_sma3nz_amean', 'mfcc3V_sma3nz_stddevNorm', 'mfcc4V_sma3nz_amean', 'mfcc4V_sma3nz_stddevNorm', 'alphaRatioUV_sma3nz_amean', 'hammarbergIndexUV_sma3nz_amean', 'slopeUV0-500_sma3nz_amean', 'slopeUV500-1500_sma3nz_amean', 'spectralFluxUV_sma3nz_amean', 'loudnessPeaksPerSec', 'VoicedSegmentsPerSec', 'MeanVoicedSegmentLengthSec', 'StddevVoicedSegmentLengthSec', 'MeanUnvoicedSegmentLength', 'StddevUnvoicedSegmentLength', 'equivalentSoundLevel_dBp']
# Drop some columns
drop_os=[
'F0semitoneFrom27.5Hz_sma3nz_amean',
'F0semitoneFrom27.5Hz_sma3nz_stddevNorm',
'F0semitoneFrom27.5Hz_sma3nz_percentile20.0',
'F0semitoneFrom27.5Hz_sma3nz_percentile50.0',
'F0semitoneFrom27.5Hz_sma3nz_percentile80.0',
'F0semitoneFrom27.5Hz_sma3nz_pctlrange0-2',
'loudness_sma3_percentile20.0',
'loudness_sma3_percentile50.0',
'loudness_sma3_percentile80.0',
'loudness_sma3_pctlrange0-2',
'HNRdBACF_sma3nz_amean',
'HNRdBACF_sma3nz_stddevNorm','logRelF0-H1-H2_sma3nz_amean',
'logRelF0-H1-H2_sma3nz_stddevNorm',
'logRelF0-H1-A3_sma3nz_amean',
'logRelF0-H1-A3_sma3nz_stddevNorm',
'F1frequency_sma3nz_amean',
'F1frequency_sma3nz_stddevNorm',
'F1bandwidth_sma3nz_amean',
'F1bandwidth_sma3nz_stddevNorm',
'F2frequency_sma3nz_amean',
'F2frequency_sma3nz_stddevNorm',
'F2bandwidth_sma3nz_amean',
'F2bandwidth_sma3nz_stddevNorm',
'F3frequency_sma3nz_amean',
'F3frequency_sma3nz_stddevNorm',
'F3bandwidth_sma3nz_amean',
'F3bandwidth_sma3nz_stddevNorm',
'spectralFlux_sma3_amean',
'spectralFlux_sma3_stddevNorm',
'mfcc1_sma3_amean',
'mfcc1_sma3_stddevNorm',
'mfcc2_sma3_amean',
'mfcc2_sma3_stddevNorm',
'mfcc3_sma3_amean',
'mfcc3_sma3_stddevNorm',
'mfcc4_sma3_amean',
'mfcc4_sma3_stddevNorm']
data.drop(drop_os, axis=1)
| participant | session | repetition | comb_no | phonation | file | Filename | start | end | H1c_mean | ... | slopeUV0-500_sma3nz_amean | slopeUV500-1500_sma3nz_amean | spectralFluxUV_sma3nz_amean | loudnessPeaksPerSec | VoicedSegmentsPerSec | MeanVoicedSegmentLengthSec | StddevVoicedSegmentLengthSec | MeanUnvoicedSegmentLength | StddevUnvoicedSegmentLength | equivalentSoundLevel_dBp | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | P0001 | S01 | R01 | S01R01 | BRT | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R01_BRT_HDM.mat | 0 days | 0 days 00:00:29.577208333 | 16.425432 | ... | -0.055642 | 0.004372 | 0.082026 | 4.228687 | 2.304304 | 0.248676 | 0.202455 | 0.171538 | 0.251367 | -38.744747 |
| 1 | P0001 | S01 | R01 | S01R01 | CRK | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R01_CRK_HDM.mat | 0 days | 0 days 00:00:27.197750 | 13.079492 | ... | -0.053462 | 0.000241 | 0.087053 | 4.819720 | 2.359012 | 0.276406 | 0.255706 | 0.128065 | 0.180544 | -41.022310 |
| 2 | P0001 | S01 | R01 | S01R01 | MOD | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R01_MOD_HDM.mat | 0 days | 0 days 00:00:26.828604167 | 13.814134 | ... | -0.059919 | 0.002522 | 0.085262 | 4.662439 | 2.017937 | 0.320370 | 0.248372 | 0.150364 | 0.215600 | -37.306293 |
| 3 | P0001 | S01 | R02 | S01R02 | BRT | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R02_BRT_HDM.mat | 0 days | 0 days 00:00:29.761750 | 14.791663 | ... | -0.042710 | -0.000428 | 0.091031 | 3.932773 | 2.356902 | 0.237143 | 0.203429 | 0.167681 | 0.222614 | -36.364433 |
| 4 | P0001 | S01 | R02 | S01R02 | CRK | /content/drive/My Drive/wp1/P0001/AC_P0001_S01... | AC_P0001_S01_R02_CRK_HDM.mat | 0 days | 0 days 00:00:28.230791667 | 14.163647 | ... | -0.055427 | -0.002846 | 0.094226 | 4.783841 | 2.094427 | 0.311186 | 0.293547 | 0.163462 | 0.249252 | -39.633247 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 112 | P0004 | S03 | R02 | S03R02 | CRK | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R02_CRK_HDM.mat | 0 days | 0 days 00:00:37.903666667 | 14.403799 | ... | -0.067100 | 0.000034 | 0.066857 | 3.299024 | 1.427061 | 0.487963 | 0.414873 | 0.215208 | 0.324859 | -36.072060 |
| 113 | P0004 | S03 | R02 | S03R02 | MOD | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R02_MOD_HDM.mat | 0 days | 0 days 00:00:33.685812500 | 16.753962 | ... | -0.054319 | 0.000623 | 0.073302 | 3.504604 | 1.516954 | 0.467059 | 0.416975 | 0.173400 | 0.266170 | -34.118713 |
| 114 | P0004 | S03 | R03 | S03R03 | BRT | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R03_BRT_HDM.mat | 0 days | 0 days 00:00:35.859729167 | 17.598709 | ... | -0.057897 | 0.001670 | 0.053314 | 3.710938 | 1.536742 | 0.422727 | 0.373083 | 0.219423 | 0.329425 | -37.928776 |
| 115 | P0004 | S03 | R03 | S03R03 | CRK | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R03_CRK_HDM.mat | 0 days | 0 days 00:00:36.052770833 | 12.654315 | ... | -0.067164 | -0.001297 | 0.060492 | 3.163152 | 1.500417 | 0.481296 | 0.399569 | 0.171961 | 0.250192 | -37.965910 |
| 116 | P0004 | S03 | R03 | S03R03 | MOD | /content/drive/My Drive/wp1/P0004/AC_P0004_S03... | AC_P0004_S03_R03_MOD_HDM.mat | 0 days | 0 days 00:00:34.081312500 | 15.900629 | ... | -0.063104 | 0.002309 | 0.068834 | 3.375404 | 1.824066 | 0.371935 | 0.310945 | 0.160167 | 0.251485 | -35.056450 |
117 rows × 117 columns
data.to_csv('/content/drive/My Drive/wp1/com_data.csv', index=None)
# full feature set
sessions = set(data['comb_no'].unique().tolist())
scores = []
f1 = []
pos_variables = []
neg_variables = []
for sp in tqdm(sessions):
# define training and test sets
train_sp = list(sessions.difference([sp]))
train_data = data[data['comb_no'].isin(train_sp)]
test_data = data[data['comb_no']==sp]
# Encoding of labels
Y_train = train_data['phonation'].tolist()
le = LabelEncoder()
Y_train = le.fit_transform(Y_train)
X_train = train_data.iloc[:,9:117]
Y_test = test_data['phonation'].tolist()
Y_test = le.transform(Y_test)
X_test = test_data.iloc[:,9:117]
# Standardize features by removing the mean and scaling to unit variance.
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
lr = LogisticRegression(penalty='l1',solver='saga',multi_class='multinomial',C=10,max_iter=2000)
lr.fit(X_train_scaled,Y_train)
X_test = scaler.transform(X_test)
scores.append(lr.score(X_test,Y_test))
preds = lr.predict(X_test)
f1.append(f1_score(Y_test,preds,average='macro'))
idx = np.argsort(lr.coef_[0])
pos_variables.append((X_train.columns[idx[-10:]].tolist(),np.sort(lr.coef_[0][idx[-10:]])))
neg_variables.append((X_train.columns[idx[0:10]].tolist(),np.sort(lr.coef_[0][idx[0:10]])))
0%| | 0/12 [00:00<?, ?it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 8%|▊ | 1/12 [00:00<00:09, 1.14it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 17%|█▋ | 2/12 [00:01<00:08, 1.23it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 25%|██▌ | 3/12 [00:02<00:07, 1.26it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 33%|███▎ | 4/12 [00:03<00:06, 1.28it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 42%|████▏ | 5/12 [00:03<00:05, 1.29it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 50%|█████ | 6/12 [00:05<00:05, 1.10it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 58%|█████▊ | 7/12 [00:06<00:05, 1.09s/it]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 67%|██████▋ | 8/12 [00:07<00:04, 1.16s/it]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 75%|███████▌ | 9/12 [00:08<00:03, 1.14s/it]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 83%|████████▎ | 10/12 [00:09<00:02, 1.02s/it]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 92%|█████████▏| 11/12 [00:10<00:00, 1.05it/s]/usr/local/lib/python3.8/dist-packages/sklearn/linear_model/_sag.py:352: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge warnings.warn( 100%|██████████| 12/12 [00:11<00:00, 1.06it/s]
print(np.mean(scores))
print(np.mean(f1))
1.0 1.0
sns.heatmap(X_train.corr(), cmap = 'bwr')
<matplotlib.axes._subplots.AxesSubplot at 0x7f2b8fd37910>
pos_variables
[(['F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H2H4c_sd',
'H1H2c_mean',
'sB1_sd',
'H1A1c_mean',
'H1A2c_sd',
'H1A1c_sd',
'H1H2c_sd',
'sF1_sd',
'sB1_mean'],
array([0.14946451, 0.18535162, 0.22830846, 0.30153633, 0.44346219,
0.47370005, 0.47875329, 0.54275334, 0.54506196, 0.95653472])),
(['H2H4c_sd',
'F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H1H2c_mean',
'sB1_sd',
'H1A2c_sd',
'H1A1c_sd',
'sF1_sd',
'H1A1c_mean',
'H1H2c_sd',
'sB1_mean'],
array([0.14659837, 0.21434797, 0.22780845, 0.22961591, 0.35466906,
0.39544233, 0.49070678, 0.50892153, 0.59053911, 0.90107841])),
(['F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H2H4c_sd',
'H1H2c_mean',
'sB1_sd',
'H1A1c_sd',
'H1A1c_mean',
'H1A2c_sd',
'sF1_sd',
'H1H2c_sd',
'sB1_mean'],
array([0.20052312, 0.23586722, 0.24429068, 0.32015241, 0.42255104,
0.46246004, 0.46714543, 0.49375374, 0.53203742, 0.92908039])),
(['H2H4c_sd',
'F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H1H2c_mean',
'sB1_sd',
'H1A1c_sd',
'H1A2c_sd',
'H1A1c_mean',
'sF1_sd',
'H1H2c_sd',
'sB1_mean'],
array([0.12940335, 0.21061938, 0.25402243, 0.30212795, 0.39196084,
0.44409866, 0.45244431, 0.5051962 , 0.50925846, 1.00676417])),
(['F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H1H2c_mean',
'H2H4c_sd',
'sB1_sd',
'H1A1c_sd',
'H1A1c_mean',
'H1A2c_sd',
'sF1_sd',
'H1H2c_sd',
'sB1_mean'],
array([0.16254657, 0.22430114, 0.2299621 , 0.33169999, 0.43263302,
0.46913753, 0.49540659, 0.49817825, 0.58556402, 0.94881636])),
(['F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H2H4c_sd',
'H1H2c_mean',
'sB1_sd',
'H1A1c_sd',
'H1A2c_sd',
'H1A1c_mean',
'sF1_sd',
'H1H2c_sd',
'sB1_mean'],
array([0.1551685 , 0.21119545, 0.27365702, 0.30148356, 0.38582488,
0.4354942 , 0.47459386, 0.54209292, 0.54424324, 0.98390506])),
(['H1H2c_mean',
'F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H2H4c_sd',
'sB1_sd',
'H1A1c_mean',
'H1A2c_sd',
'H1A1c_sd',
'sF1_sd',
'H1H2c_sd',
'sB1_mean'],
array([0.19051858, 0.19131472, 0.2868289 , 0.33627055, 0.41539939,
0.47924503, 0.50136866, 0.64646879, 0.78800712, 0.88059084])),
(['F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H2H4c_sd',
'H1H2c_mean',
'sB1_sd',
'H1A1c_sd',
'H1A1c_mean',
'sF1_sd',
'H1A2c_sd',
'H1H2c_sd',
'sB1_mean'],
array([0.09984515, 0.20419674, 0.24524062, 0.31151171, 0.41387609,
0.45171565, 0.54143803, 0.55204479, 0.59520872, 0.94110268])),
(['H1c_mean',
'H2H4c_sd',
'H1H2c_mean',
'sB1_sd',
'H1A2c_sd',
'H1H2c_sd',
'H1A1c_sd',
'H1A1c_mean',
'sF1_sd',
'sB1_mean'],
array([0.04740312, 0.14981009, 0.1742715 , 0.23119869, 0.48213027,
0.49781757, 0.49885554, 0.51295122, 0.66697745, 1.08975681])),
(['H1c_mean',
'F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H1H2c_mean',
'sB1_sd',
'H1A1c_sd',
'H1A2c_sd',
'H1A1c_mean',
'H1H2c_sd',
'sF1_sd',
'sB1_mean'],
array([0.06666335, 0.17528665, 0.26081346, 0.26531245, 0.44717935,
0.47692868, 0.48617929, 0.60882136, 0.63587393, 0.9510385 ])),
(['H2H4c_sd',
'F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H1H2c_mean',
'sB1_sd',
'H1A1c_mean',
'H1A1c_sd',
'H1A2c_sd',
'sF1_sd',
'H1H2c_sd',
'sB1_mean'],
array([0.15705978, 0.16335742, 0.22772005, 0.29078403, 0.45132345,
0.45836308, 0.47413699, 0.52352131, 0.54942839, 0.96057728])),
(['H2H4c_sd',
'F1amplitudeLogRelF0_sma3nz_stddevNorm',
'H1H2c_mean',
'sB1_sd',
'H1A1c_mean',
'H1H2c_sd',
'H1A2c_sd',
'H1A1c_sd',
'sF1_sd',
'sB1_mean'],
array([0.08373586, 0.19288593, 0.19498619, 0.26920201, 0.43770287,
0.44239404, 0.47793933, 0.49576434, 0.63695295, 0.91319489]))]
neg_variables
[(['mfcc1_sma3_amean',
'sF2_sd',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'HNR05_sd',
'shimmerLocaldB_sma3nz_amean',
'sF0_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'sB2_sd',
'loudness_sma3_meanRisingSlope',
'HNR35_mean'],
array([-0.44297849, -0.39240058, -0.33170551, -0.28845043, -0.25007928,
-0.18447656, -0.15340135, -0.0143297 , -0.01072748, -0.00951651])),
(['sF2_sd',
'mfcc1_sma3_amean',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'shimmerLocaldB_sma3nz_amean',
'F1amplitudeLogRelF0_sma3nz_amean',
'sF0_sd',
'HNR05_sd',
'loudness_sma3_meanRisingSlope',
'sF4_sd',
'loudness_sma3_stddevRisingSlope'],
array([-0.45844992, -0.45073872, -0.43184895, -0.2418072 , -0.21037039,
-0.19272285, -0.13255871, -0.08374159, 0. , 0. ])),
(['mfcc1_sma3_amean',
'sF2_sd',
'HNR05_sd',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'shimmerLocaldB_sma3nz_amean',
'F1amplitudeLogRelF0_sma3nz_amean',
'sF0_sd',
'HNR35_mean',
'sB2_sd',
'F0semitoneFrom27.5Hz_sma3nz_percentile20.0'],
array([-0.47364353, -0.37937342, -0.3383721 , -0.2668099 , -0.21952428,
-0.1938352 , -0.17663745, -0.01486526, -0.00871667, 0. ])),
(['mfcc1_sma3_amean',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'sF2_sd',
'HNR05_sd',
'sF0_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'shimmerLocaldB_sma3nz_amean',
'sB2_sd',
'loudness_sma3_stddevRisingSlope',
'loudness_sma3_meanRisingSlope'],
array([-0.46740037, -0.42326165, -0.33321037, -0.30196819, -0.23079317,
-0.22045629, -0.13416685, -0.02633928, 0. , 0. ])),
(['mfcc1_sma3_amean',
'sF2_sd',
'shimmerLocaldB_sma3nz_amean',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'HNR05_sd',
'sF0_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'loudness_sma3_meanRisingSlope',
'sF4_sd',
'loudness_sma3_stddevRisingSlope'],
array([-0.53438737, -0.3478746 , -0.32682659, -0.30183222, -0.27813303,
-0.24856659, -0.17461055, -0.00419015, 0. , 0. ])),
(['mfcc1_sma3_amean',
'sF2_sd',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'shimmerLocaldB_sma3nz_amean',
'HNR05_sd',
'sF0_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'loudness_sma3_meanRisingSlope',
'HNR35_mean',
'loudness_sma3_stddevRisingSlope'],
array([-0.4289039 , -0.33627416, -0.30029402, -0.26534171, -0.22575912,
-0.1695599 , -0.13458875, -0.0611361 , -0.00113178, 0. ])),
(['mfcc1_sma3_amean',
'sF2_sd',
'HNR05_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'sF0_sd',
'shimmerLocaldB_sma3nz_amean',
'loudness_sma3_meanRisingSlope',
'sF4_sd',
'loudness_sma3_stddevRisingSlope'],
array([-0.41152451, -0.31307196, -0.18461179, -0.183275 , -0.1154726 ,
-0.06720962, -0.06580016, -0.01441295, 0. , 0. ])),
(['mfcc1_sma3_amean',
'sF2_sd',
'shimmerLocaldB_sma3nz_amean',
'HNR05_sd',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'sF0_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'loudness_sma3_meanRisingSlope',
'sB2_sd',
'loudness_sma3_stddevRisingSlope'],
array([-0.43641105, -0.39286968, -0.25257285, -0.23983572, -0.21799812,
-0.19812325, -0.08517395, -0.02398583, -0.02335311, 0. ])),
(['mfcc1_sma3_amean',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'sF2_sd',
'HNR05_sd',
'sF0_sd',
'shimmerLocaldB_sma3nz_amean',
'loudness_sma3_stddevFallingSlope',
'loudness_sma3_meanFallingSlope',
'loudness_sma3_stddevRisingSlope',
'loudness_sma3_meanRisingSlope'],
array([-0.54719593, -0.3528045 , -0.31144928, -0.27336489, -0.25635667,
-0.14040009, 0. , 0. , 0. , 0. ])),
(['mfcc1_sma3_amean',
'sF2_sd',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'F1amplitudeLogRelF0_sma3nz_amean',
'HNR05_sd',
'shimmerLocaldB_sma3nz_amean',
'sF0_sd',
'HNR35_mean',
'loudness_sma3_stddevRisingSlope',
'loudness_sma3_meanRisingSlope'],
array([-0.44330906, -0.43186172, -0.24570397, -0.18320193, -0.16888428,
-0.15553506, -0.13670048, -0.04547549, 0. , 0. ])),
(['mfcc1_sma3_amean',
'sF2_sd',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'HNR05_sd',
'shimmerLocaldB_sma3nz_amean',
'sF0_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'HNR35_mean',
'loudness_sma3_meanRisingSlope',
'loudness_sma3_stddevRisingSlope'],
array([-0.48224246, -0.37631924, -0.30397032, -0.28733498, -0.27091547,
-0.17064388, -0.16160793, -0.01347451, -0.01255486, 0. ])),
(['mfcc1_sma3_amean',
'F0semitoneFrom27.5Hz_sma3nz_meanFallingSlope',
'sF2_sd',
'shimmerLocaldB_sma3nz_amean',
'HNR05_sd',
'F1amplitudeLogRelF0_sma3nz_amean',
'sF3_sd',
'sF0_sd',
'loudness_sma3_meanRisingSlope',
'sB2_sd'],
array([-0.50528533, -0.35721019, -0.31525785, -0.26432352, -0.25706794,
-0.20611342, -0.0920233 , -0.0738398 , -0.04582519, -0.00526275]))]
The data size is too small for combined feature set. The following part is not analysed until I have more data.